data_source
stringclasses 1
value | reward_model
dict | extra_info
dict | prompt
listlengths 2
2
| ability
stringclasses 1
value | env_name
stringclasses 1
value | images
images listlengths 1
1
|
|---|---|---|---|---|---|---|
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the tissue box is blue."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_39518.jpg",
"item_id": "sa_39518",
"options": [
"The color of the tissue box is blue.",
"The color of the tissue box is gray.",
"The color of the tissue box is black.",
"The color of the tissue box is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the tissue box?\nOptions: \nA. The color of the tissue box is blue.\nB. The color of the tissue box is gray.\nC. The color of the tissue box is black.\nD. The color of the tissue box is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the tissue box?\nOptions: \nA. The color of the tissue box is blue.\nB. The color of the tissue box is gray.\nC. The color of the tissue box is black.\nD. The color of the tissue box is white.\n\nOptions:\nA. The color of the tissue box is blue.\nB. The color of the tissue box is gray.\nC. The color of the tissue box is black.\nD. The color of the tissue box is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the bucket is white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_50451.jpg",
"item_id": "sa_50451",
"options": [
"The color of the bucket is white.",
"The color of the bucket is black.",
"The color of the bucket is green.",
"The color of the bucket is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the bucket?\nOptions: \nA. The color of the bucket is white.\nB. The color of the bucket is black.\nC. The color of the bucket is green.\nD. The color of the bucket is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the bucket?\nOptions: \nA. The color of the bucket is white.\nB. The color of the bucket is black.\nC. The color of the bucket is green.\nD. The color of the bucket is red.\n\nOptions:\nA. The color of the bucket is white.\nB. The color of the bucket is black.\nC. The color of the bucket is green.\nD. The color of the bucket is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the truck is yellow."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_17108.jpg",
"item_id": "sa_17108",
"options": [
"The color of the truck is yellow.",
"The color of the truck is white.",
"The color of the truck is silver.",
"The color of the truck is black."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the truck?\nOptions: \nA. The color of the truck is yellow.\nB. The color of the truck is white.\nC. The color of the truck is silver.\nD. The color of the truck is black.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the truck?\nOptions: \nA. The color of the truck is yellow.\nB. The color of the truck is white.\nC. The color of the truck is silver.\nD. The color of the truck is black.\n\nOptions:\nA. The color of the truck is yellow.\nB. The color of the truck is white.\nC. The color of the truck is silver.\nD. The color of the truck is black.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the shovel is blue."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_65370.jpg",
"item_id": "sa_65370",
"options": [
"The color of the shovel is blue.",
"The color of the shovel is black.",
"The color of the shovel is red.",
"The color of the shovel is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the shovel?\nOptions: \nA. The color of the shovel is blue.\nB. The color of the shovel is black.\nC. The color of the shovel is red.\nD. The color of the shovel is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the shovel?\nOptions: \nA. The color of the shovel is blue.\nB. The color of the shovel is black.\nC. The color of the shovel is red.\nD. The color of the shovel is yellow.\n\nOptions:\nA. The color of the shovel is blue.\nB. The color of the shovel is black.\nC. The color of the shovel is red.\nD. The color of the shovel is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the watchband is purple."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_11405.jpg",
"item_id": "sa_11405",
"options": [
"The color of the watchband is purple.",
"The color of the watchband is black.",
"The color of the watchband is blue.",
"The color of the watchband is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the watchband?\nOptions: \nA. The color of the watchband is purple.\nB. The color of the watchband is black.\nC. The color of the watchband is blue.\nD. The color of the watchband is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the watchband?\nOptions: \nA. The color of the watchband is purple.\nB. The color of the watchband is black.\nC. The color of the watchband is blue.\nD. The color of the watchband is red.\n\nOptions:\nA. The color of the watchband is purple.\nB. The color of the watchband is black.\nC. The color of the watchband is blue.\nD. The color of the watchband is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The animal in the poster is a dove."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10997.jpg",
"item_id": "sa_10997",
"options": [
"The animal in the poster is a dove.",
"The animal in the poster is a cat.",
"The animal in the poster is a horse.",
"The animal in the poster is a dog."
],
"prompt_preset": "seal_zoom",
"question": "Question: What kind of animal is in the poster?\nOptions: \nA. The animal in the poster is a dove.\nB. The animal in the poster is a cat.\nC. The animal in the poster is a horse.\nD. The animal in the poster is a dog.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What kind of animal is in the poster?\nOptions: \nA. The animal in the poster is a dove.\nB. The animal in the poster is a cat.\nC. The animal in the poster is a horse.\nD. The animal in the poster is a dog.\n\nOptions:\nA. The animal in the poster is a dove.\nB. The animal in the poster is a cat.\nC. The animal in the poster is a horse.\nD. The animal in the poster is a dog.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the helmet is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_25212.jpg",
"item_id": "sa_25212",
"options": [
"The color of the helmet is red.",
"The color of the helmet is white.",
"The color of the helmet is yellow.",
"The color of the helmet is blue."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the helmet?\nOptions: \nA. The color of the helmet is red.\nB. The color of the helmet is white.\nC. The color of the helmet is yellow.\nD. The color of the helmet is blue.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the helmet?\nOptions: \nA. The color of the helmet is red.\nB. The color of the helmet is white.\nC. The color of the helmet is yellow.\nD. The color of the helmet is blue.\n\nOptions:\nA. The color of the helmet is red.\nB. The color of the helmet is white.\nC. The color of the helmet is yellow.\nD. The color of the helmet is blue.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the bus is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_7065.jpg",
"item_id": "sa_7065",
"options": [
"The color of the bus is red.",
"The color of the bus is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: Is the color of the bus red or yellow?\nOptions: \nA. The color of the bus is red.\nB. The color of the bus is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: Is the color of the bus red or yellow?\nOptions: \nA. The color of the bus is red.\nB. The color of the bus is yellow.\n\nOptions:\nA. The color of the bus is red.\nB. The color of the bus is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the trash can is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_29509.jpg",
"item_id": "sa_29509",
"options": [
"The color of the trash can is black.",
"The color of the trash can is green.",
"The color of the trash can is yellow.",
"The color of the trash can is silver."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the trash can?\nOptions: \nA. The color of the trash can is black.\nB. The color of the trash can is green.\nC. The color of the trash can is yellow.\nD. The color of the trash can is silver.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the trash can?\nOptions: \nA. The color of the trash can is black.\nB. The color of the trash can is green.\nC. The color of the trash can is yellow.\nD. The color of the trash can is silver.\n\nOptions:\nA. The color of the trash can is black.\nB. The color of the trash can is green.\nC. The color of the trash can is yellow.\nD. The color of the trash can is silver.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the candles is yellow."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_9019.jpg",
"item_id": "sa_9019",
"options": [
"The color of the candles is yellow.",
"The color of the candles is white.",
"The color of the candles is red.",
"The color of the candles is gray."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the candles?\nOptions: \nA. The color of the candles is yellow.\nB. The color of the candles is white.\nC. The color of the candles is red.\nD. The color of the candles is gray.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the candles?\nOptions: \nA. The color of the candles is yellow.\nB. The color of the candles is white.\nC. The color of the candles is red.\nD. The color of the candles is gray.\n\nOptions:\nA. The color of the candles is yellow.\nB. The color of the candles is white.\nC. The color of the candles is red.\nD. The color of the candles is gray.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the life buoy is red and white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_27986.jpg",
"item_id": "sa_27986",
"options": [
"The color of the life buoy is red and white.",
"The color of the life buoy is blue and white.",
"The color of the life buoy is green and white.",
"The color of the life buoy is yellow and red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the life buoy?\nOptions: \nA. The color of the life buoy is red and white.\nB. The color of the life buoy is blue and white.\nC. The color of the life buoy is green and white.\nD. The color of the life buoy is yellow and red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the life buoy?\nOptions: \nA. The color of the life buoy is red and white.\nB. The color of the life buoy is blue and white.\nC. The color of the life buoy is green and white.\nD. The color of the life buoy is yellow and red.\n\nOptions:\nA. The color of the life buoy is red and white.\nB. The color of the life buoy is blue and white.\nC. The color of the life buoy is green and white.\nD. The color of the life buoy is yellow and red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the woman's handbag is brown."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_38669.jpg",
"item_id": "sa_38669",
"options": [
"The color of the woman's handbag is brown.",
"The color of the woman's handbag is black.",
"The color of the woman's handbag is white.",
"The color of the woman's handbag is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the woman's handbag?\nOptions: \nA. The color of the woman's handbag is brown.\nB. The color of the woman's handbag is black.\nC. The color of the woman's handbag is white.\nD. The color of the woman's handbag is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the woman's handbag?\nOptions: \nA. The color of the woman's handbag is brown.\nB. The color of the woman's handbag is black.\nC. The color of the woman's handbag is white.\nD. The color of the woman's handbag is red.\n\nOptions:\nA. The color of the woman's handbag is brown.\nB. The color of the woman's handbag is black.\nC. The color of the woman's handbag is white.\nD. The color of the woman's handbag is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the lock is silver."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_70112.jpg",
"item_id": "sa_70112",
"options": [
"The color of the lock is silver.",
"The color of the lock is red.",
"The color of the lock is yellow.",
"The color of the lock is golden."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the lock?\nOptions: \nA. The color of the lock is silver.\nB. The color of the lock is red.\nC. The color of the lock is yellow.\nD. The color of the lock is golden.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the lock?\nOptions: \nA. The color of the lock is silver.\nB. The color of the lock is red.\nC. The color of the lock is yellow.\nD. The color of the lock is golden.\n\nOptions:\nA. The color of the lock is silver.\nB. The color of the lock is red.\nC. The color of the lock is yellow.\nD. The color of the lock is golden.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the motorcycle is orange."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_6422.jpg",
"item_id": "sa_6422",
"options": [
"The color of the motorcycle is orange.",
"The color of the motorcycle is green."
],
"prompt_preset": "seal_zoom",
"question": "Question: Is the color of the motorcycle orange or green?\nOptions: \nA. The color of the motorcycle is orange.\nB. The color of the motorcycle is green.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: Is the color of the motorcycle orange or green?\nOptions: \nA. The color of the motorcycle is orange.\nB. The color of the motorcycle is green.\n\nOptions:\nA. The color of the motorcycle is orange.\nB. The color of the motorcycle is green.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the hat is white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_31451.jpg",
"item_id": "sa_31451",
"options": [
"The color of the hat is white.",
"The color of the hat is black.",
"The color of the hat is pink.",
"The color of the hat is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the hat?\nOptions: \nA. The color of the hat is white.\nB. The color of the hat is black.\nC. The color of the hat is pink.\nD. The color of the hat is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the hat?\nOptions: \nA. The color of the hat is white.\nB. The color of the hat is black.\nC. The color of the hat is pink.\nD. The color of the hat is yellow.\n\nOptions:\nA. The color of the hat is white.\nB. The color of the hat is black.\nC. The color of the hat is pink.\nD. The color of the hat is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the plastic bucket is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_9957.jpg",
"item_id": "sa_9957",
"options": [
"The color of the plastic bucket is green.",
"The color of the plastic bucket is black."
],
"prompt_preset": "seal_zoom",
"question": "Question: Is the color of the plastic bucket black or green?\nOptions: \nA. The color of the plastic bucket is green.\nB. The color of the plastic bucket is black.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: Is the color of the plastic bucket black or green?\nOptions: \nA. The color of the plastic bucket is green.\nB. The color of the plastic bucket is black.\n\nOptions:\nA. The color of the plastic bucket is green.\nB. The color of the plastic bucket is black.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the earring is blue."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_11370.jpg",
"item_id": "sa_11370",
"options": [
"The color of the earring is blue.",
"The color of the earring is pink.",
"The color of the earring is silver.",
"The color of the earring is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the earring?\nOptions: \nA. The color of the earring is blue.\nB. The color of the earring is pink.\nC. The color of the earring is silver.\nD. The color of the earring is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the earring?\nOptions: \nA. The color of the earring is blue.\nB. The color of the earring is pink.\nC. The color of the earring is silver.\nD. The color of the earring is red.\n\nOptions:\nA. The color of the earring is blue.\nB. The color of the earring is pink.\nC. The color of the earring is silver.\nD. The color of the earring is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the straw is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_68625.jpg",
"item_id": "sa_68625",
"options": [
"The color of the straw is black.",
"The color of the straw is white.",
"The color of the straw is red.",
"The color of the straw is pink."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the straw?\nOptions: \nA. The color of the straw is black.\nB. The color of the straw is white.\nC. The color of the straw is red.\nD. The color of the straw is pink.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the straw?\nOptions: \nA. The color of the straw is black.\nB. The color of the straw is white.\nC. The color of the straw is red.\nD. The color of the straw is pink.\n\nOptions:\nA. The color of the straw is black.\nB. The color of the straw is white.\nC. The color of the straw is red.\nD. The color of the straw is pink.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the man's cap is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_6965.jpg",
"item_id": "sa_6965",
"options": [
"The color of the man's cap is red.",
"The color of the man's cap is blue.",
"The color of the man's cap is white.",
"The color of the man's cap is gray."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the man's cap?\nOptions: \nA. The color of the man's cap is red.\nB. The color of the man's cap is blue.\nC. The color of the man's cap is white.\nD. The color of the man's cap is gray.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the man's cap?\nOptions: \nA. The color of the man's cap is red.\nB. The color of the man's cap is blue.\nC. The color of the man's cap is white.\nD. The color of the man's cap is gray.\n\nOptions:\nA. The color of the man's cap is red.\nB. The color of the man's cap is blue.\nC. The color of the man's cap is white.\nD. The color of the man's cap is gray.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the helmet is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_39678.jpg",
"item_id": "sa_39678",
"options": [
"The color of the helmet is green.",
"The color of the helmet is blue.",
"The color of the helmet is white.",
"The color of the helmet is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the helmet?\nOptions: \nA. The color of the helmet is green.\nB. The color of the helmet is blue.\nC. The color of the helmet is white.\nD. The color of the helmet is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the helmet?\nOptions: \nA. The color of the helmet is green.\nB. The color of the helmet is blue.\nC. The color of the helmet is white.\nD. The color of the helmet is red.\n\nOptions:\nA. The color of the helmet is green.\nB. The color of the helmet is blue.\nC. The color of the helmet is white.\nD. The color of the helmet is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the mailbox is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_86882.jpg",
"item_id": "sa_86882",
"options": [
"The color of the mailbox is black.",
"The color of the mailbox is red.",
"The color of the mailbox is yellow.",
"The color of the mailbox is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the mailbox?\nOptions: \nA. The color of the mailbox is black.\nB. The color of the mailbox is red.\nC. The color of the mailbox is yellow.\nD. The color of the mailbox is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the mailbox?\nOptions: \nA. The color of the mailbox is black.\nB. The color of the mailbox is red.\nC. The color of the mailbox is yellow.\nD. The color of the mailbox is white.\n\nOptions:\nA. The color of the mailbox is black.\nB. The color of the mailbox is red.\nC. The color of the mailbox is yellow.\nD. The color of the mailbox is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the backpack is gray."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_28392.jpg",
"item_id": "sa_28392",
"options": [
"The color of the backpack is gray.",
"The color of the backpack is yellow.",
"The color of the backpack is white.",
"The color of the backpack is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the backpack?\nOptions: \nA. The color of the backpack is gray.\nB. The color of the backpack is yellow.\nC. The color of the backpack is white.\nD. The color of the backpack is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the backpack?\nOptions: \nA. The color of the backpack is gray.\nB. The color of the backpack is yellow.\nC. The color of the backpack is white.\nD. The color of the backpack is red.\n\nOptions:\nA. The color of the backpack is gray.\nB. The color of the backpack is yellow.\nC. The color of the backpack is white.\nD. The color of the backpack is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the woman's dress is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_19272.jpg",
"item_id": "sa_19272",
"options": [
"The color of the woman's dress is black.",
"The color of the woman's dress is pink.",
"The color of the woman's dress is white.",
"The color of the woman's dress is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the woman's dress?\nOptions: \nA. The color of the woman's dress is black.\nB. The color of the woman's dress is pink.\nC. The color of the woman's dress is white.\nD. The color of the woman's dress is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the woman's dress?\nOptions: \nA. The color of the woman's dress is black.\nB. The color of the woman's dress is pink.\nC. The color of the woman's dress is white.\nD. The color of the woman's dress is red.\n\nOptions:\nA. The color of the woman's dress is black.\nB. The color of the woman's dress is pink.\nC. The color of the woman's dress is white.\nD. The color of the woman's dress is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the plastic basin is pink."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10052.jpg",
"item_id": "sa_10052",
"options": [
"The color of the plastic basin is pink.",
"The color of the plastic basin is yellow.",
"The color of the plastic basin is blue.",
"The color of the plastic basin is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the plastic basin?\nOptions: \nA. The color of the plastic basin is pink.\nB. The color of the plastic basin is yellow.\nC. The color of the plastic basin is blue.\nD. The color of the plastic basin is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the plastic basin?\nOptions: \nA. The color of the plastic basin is pink.\nB. The color of the plastic basin is yellow.\nC. The color of the plastic basin is blue.\nD. The color of the plastic basin is white.\n\nOptions:\nA. The color of the plastic basin is pink.\nB. The color of the plastic basin is yellow.\nC. The color of the plastic basin is blue.\nD. The color of the plastic basin is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the flag is blue and yellow."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10885.jpg",
"item_id": "sa_10885",
"options": [
"The color of the flag is blue and yellow.",
"The color of the flag is red and yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: Is the flag blue and yellow or red and yellow?\nOptions: \nA. The color of the flag is blue and yellow.\nB. The color of the flag is red and yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: Is the flag blue and yellow or red and yellow?\nOptions: \nA. The color of the flag is blue and yellow.\nB. The color of the flag is red and yellow.\n\nOptions:\nA. The color of the flag is blue and yellow.\nB. The color of the flag is red and yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the broom is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_80880.jpg",
"item_id": "sa_80880",
"options": [
"The color of the broom is black.",
"The color of the broom is yellow.",
"The color of the broom is red.",
"The color of the broom is gray."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the broom?\nOptions: \nA. The color of the broom is black.\nB. The color of the broom is yellow.\nC. The color of the broom is red.\nD. The color of the broom is gray.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the broom?\nOptions: \nA. The color of the broom is black.\nB. The color of the broom is yellow.\nC. The color of the broom is red.\nD. The color of the broom is gray.\n\nOptions:\nA. The color of the broom is black.\nB. The color of the broom is yellow.\nC. The color of the broom is red.\nD. The color of the broom is gray.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the bicycle is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_9230.jpg",
"item_id": "sa_9230",
"options": [
"The color of the bicycle is red.",
"The color of the bicycle is blue.",
"The color of the bicycle is white.",
"The color of the bicycle is silver."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the bicycle?\nOptions: \nA. The color of the bicycle is red.\nB. The color of the bicycle is blue.\nC. The color of the bicycle is white.\nD. The color of the bicycle is silver.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the bicycle?\nOptions: \nA. The color of the bicycle is red.\nB. The color of the bicycle is blue.\nC. The color of the bicycle is white.\nD. The color of the bicycle is silver.\n\nOptions:\nA. The color of the bicycle is red.\nB. The color of the bicycle is blue.\nC. The color of the bicycle is white.\nD. The color of the bicycle is silver.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the tent is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_7668.jpg",
"item_id": "sa_7668",
"options": [
"The color of the tent is green.",
"The color of the tent is blue.",
"The color of the tent is white.",
"The color of the tent is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the tent?\nOptions: \nA. The color of the tent is green.\nB. The color of the tent is blue.\nC. The color of the tent is white.\nD. The color of the tent is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the tent?\nOptions: \nA. The color of the tent is green.\nB. The color of the tent is blue.\nC. The color of the tent is white.\nD. The color of the tent is yellow.\n\nOptions:\nA. The color of the tent is green.\nB. The color of the tent is blue.\nC. The color of the tent is white.\nD. The color of the tent is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the messenger bag is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_16825.jpg",
"item_id": "sa_16825",
"options": [
"The color of the messenger bag is black.",
"The color of the messenger bag is brown.",
"The color of the messenger bag is red.",
"The color of the messenger bag is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the messenger bag?\nOptions: \nA. The color of the messenger bag is black.\nB. The color of the messenger bag is brown.\nC. The color of the messenger bag is red.\nD. The color of the messenger bag is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the messenger bag?\nOptions: \nA. The color of the messenger bag is black.\nB. The color of the messenger bag is brown.\nC. The color of the messenger bag is red.\nD. The color of the messenger bag is white.\n\nOptions:\nA. The color of the messenger bag is black.\nB. The color of the messenger bag is brown.\nC. The color of the messenger bag is red.\nD. The color of the messenger bag is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the motorcycle is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_66699.jpg",
"item_id": "sa_66699",
"options": [
"The color of the motorcycle is black.",
"The color of the motorcycle is red.",
"The color of the motorcycle is orange.",
"The color of the motorcycle is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the motorcycle?\nOptions: \nA. The color of the motorcycle is black.\nB. The color of the motorcycle is red.\nC. The color of the motorcycle is orange.\nD. The color of the motorcycle is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the motorcycle?\nOptions: \nA. The color of the motorcycle is black.\nB. The color of the motorcycle is red.\nC. The color of the motorcycle is orange.\nD. The color of the motorcycle is white.\n\nOptions:\nA. The color of the motorcycle is black.\nB. The color of the motorcycle is red.\nC. The color of the motorcycle is orange.\nD. The color of the motorcycle is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the car is white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_43391.jpg",
"item_id": "sa_43391",
"options": [
"The color of the car is white.",
"The color of the car is black.",
"The color of the car is silver.",
"The color of the car is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the car?\nOptions: \nA. The color of the car is white.\nB. The color of the car is black.\nC. The color of the car is silver.\nD. The color of the car is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the car?\nOptions: \nA. The color of the car is white.\nB. The color of the car is black.\nC. The color of the car is silver.\nD. The color of the car is red.\n\nOptions:\nA. The color of the car is white.\nB. The color of the car is black.\nC. The color of the car is silver.\nD. The color of the car is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the fire hydrant is yellow."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10415.jpg",
"item_id": "sa_10415",
"options": [
"The color of the fire hydrant is yellow.",
"The color of the fire hydrant is red.",
"The color of the fire hydrant is black.",
"The color of the fire hydrant is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the fire hydrant?\nOptions: \nA. The color of the fire hydrant is yellow.\nB. The color of the fire hydrant is red.\nC. The color of the fire hydrant is black.\nD. The color of the fire hydrant is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the fire hydrant?\nOptions: \nA. The color of the fire hydrant is yellow.\nB. The color of the fire hydrant is red.\nC. The color of the fire hydrant is black.\nD. The color of the fire hydrant is white.\n\nOptions:\nA. The color of the fire hydrant is yellow.\nB. The color of the fire hydrant is red.\nC. The color of the fire hydrant is black.\nD. The color of the fire hydrant is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the cyclist's box is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_25835.jpg",
"item_id": "sa_25835",
"options": [
"The color of the cyclist's box is green.",
"The color of the cyclist's box is black.",
"The color of the cyclist's box is yellow.",
"The color of the cyclist's box is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the cyclist's box?\nOptions: \nA. The color of the cyclist's box is green.\nB. The color of the cyclist's box is black.\nC. The color of the cyclist's box is yellow.\nD. The color of the cyclist's box is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the cyclist's box?\nOptions: \nA. The color of the cyclist's box is green.\nB. The color of the cyclist's box is black.\nC. The color of the cyclist's box is yellow.\nD. The color of the cyclist's box is red.\n\nOptions:\nA. The color of the cyclist's box is green.\nB. The color of the cyclist's box is black.\nC. The color of the cyclist's box is yellow.\nD. The color of the cyclist's box is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the pet collar is purple."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_52155.jpg",
"item_id": "sa_52155",
"options": [
"The color of the pet collar is purple.",
"The color of the pet collar is black.",
"The color of the pet collar is red.",
"The color of the pet collar is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the pet collar?\nOptions: \nA. The color of the pet collar is purple.\nB. The color of the pet collar is black.\nC. The color of the pet collar is red.\nD. The color of the pet collar is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the pet collar?\nOptions: \nA. The color of the pet collar is purple.\nB. The color of the pet collar is black.\nC. The color of the pet collar is red.\nD. The color of the pet collar is yellow.\n\nOptions:\nA. The color of the pet collar is purple.\nB. The color of the pet collar is black.\nC. The color of the pet collar is red.\nD. The color of the pet collar is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the handbag is yellow."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_31511.jpg",
"item_id": "sa_31511",
"options": [
"The color of the handbag is yellow.",
"The color of the handbag is white.",
"The color of the handbag is red.",
"The color of the handbag is black."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the handbag?\nOptions: \nA. The color of the handbag is yellow.\nB. The color of the handbag is white.\nC. The color of the handbag is red.\nD. The color of the handbag is black.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the handbag?\nOptions: \nA. The color of the handbag is yellow.\nB. The color of the handbag is white.\nC. The color of the handbag is red.\nD. The color of the handbag is black.\n\nOptions:\nA. The color of the handbag is yellow.\nB. The color of the handbag is white.\nC. The color of the handbag is red.\nD. The color of the handbag is black.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The kid has curly hair."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10225.jpg",
"item_id": "sa_10225",
"options": [
"The kid has curly hair.",
"The kid has straight hair."
],
"prompt_preset": "seal_zoom",
"question": "Question: Does the kid have curly hair or straight hair?\nOptions: \nA. The kid has curly hair.\nB. The kid has straight hair.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: Does the kid have curly hair or straight hair?\nOptions: \nA. The kid has curly hair.\nB. The kid has straight hair.\n\nOptions:\nA. The kid has curly hair.\nB. The kid has straight hair.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the trash can is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_7081.jpg",
"item_id": "sa_7081",
"options": [
"The color of the trash can is green.",
"The color of the trash can is black.",
"The color of the trash can is red.",
"The color of the trash can is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the trash can?\nOptions: \nA. The color of the trash can is green.\nB. The color of the trash can is black.\nC. The color of the trash can is red.\nD. The color of the trash can is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the trash can?\nOptions: \nA. The color of the trash can is green.\nB. The color of the trash can is black.\nC. The color of the trash can is red.\nD. The color of the trash can is yellow.\n\nOptions:\nA. The color of the trash can is green.\nB. The color of the trash can is black.\nC. The color of the trash can is red.\nD. The color of the trash can is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the dustpan is blue."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_86101.jpg",
"item_id": "sa_86101",
"options": [
"The color of the dustpan is blue.",
"The color of the dustpan is red.",
"The color of the dustpan is white.",
"The color of the dustpan is purple."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the dustpan?\nOptions: \nA. The color of the dustpan is blue.\nB. The color of the dustpan is red.\nC. The color of the dustpan is white.\nD. The color of the dustpan is purple.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the dustpan?\nOptions: \nA. The color of the dustpan is blue.\nB. The color of the dustpan is red.\nC. The color of the dustpan is white.\nD. The color of the dustpan is purple.\n\nOptions:\nA. The color of the dustpan is blue.\nB. The color of the dustpan is red.\nC. The color of the dustpan is white.\nD. The color of the dustpan is purple.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The breed of the dog is golden retriever."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_20519.jpg",
"item_id": "sa_20519",
"options": [
"The breed of the dog is golden retriever.",
"The breed of the dog is Dalmatian.",
"The breed of the dog is corgi.",
"The breed of the dog is Husky."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the breed of the dog?\nOptions: \nA. The breed of the dog is golden retriever.\nB. The breed of the dog is Dalmatian.\nC. The breed of the dog is corgi.\nD. The breed of the dog is Husky.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the breed of the dog?\nOptions: \nA. The breed of the dog is golden retriever.\nB. The breed of the dog is Dalmatian.\nC. The breed of the dog is corgi.\nD. The breed of the dog is Husky.\n\nOptions:\nA. The breed of the dog is golden retriever.\nB. The breed of the dog is Dalmatian.\nC. The breed of the dog is corgi.\nD. The breed of the dog is Husky.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The material of the stool is plastic."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_9611.jpg",
"item_id": "sa_9611",
"options": [
"The material of the stool is plastic.",
"The material of the stool is wood.",
"The material of the stool is steel.",
"The material of the stool is bamboo."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the material of the stool?\nOptions: \nA. The material of the stool is plastic.\nB. The material of the stool is wood.\nC. The material of the stool is steel.\nD. The material of the stool is bamboo.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the material of the stool?\nOptions: \nA. The material of the stool is plastic.\nB. The material of the stool is wood.\nC. The material of the stool is steel.\nD. The material of the stool is bamboo.\n\nOptions:\nA. The material of the stool is plastic.\nB. The material of the stool is wood.\nC. The material of the stool is steel.\nD. The material of the stool is bamboo.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the luggage is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_19483.jpg",
"item_id": "sa_19483",
"options": [
"The color of the luggage is red.",
"The color of the luggage is yellow.",
"The color of the luggage is black.",
"The color of the luggage is green."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the luggage?\nOptions: \nA. The color of the luggage is red.\nB. The color of the luggage is yellow.\nC. The color of the luggage is black.\nD. The color of the luggage is green.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the luggage?\nOptions: \nA. The color of the luggage is red.\nB. The color of the luggage is yellow.\nC. The color of the luggage is black.\nD. The color of the luggage is green.\n\nOptions:\nA. The color of the luggage is red.\nB. The color of the luggage is yellow.\nC. The color of the luggage is black.\nD. The color of the luggage is green.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the dog is black and white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_59664.jpg",
"item_id": "sa_59664",
"options": [
"The color of the dog is black and white.",
"The color of the dog is brown.",
"The color of the dog is gray.",
"The color of the dog is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the dog?\nOptions: \nA. The color of the dog is black and white.\nB. The color of the dog is brown.\nC. The color of the dog is gray.\nD. The color of the dog is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the dog?\nOptions: \nA. The color of the dog is black and white.\nB. The color of the dog is brown.\nC. The color of the dog is gray.\nD. The color of the dog is yellow.\n\nOptions:\nA. The color of the dog is black and white.\nB. The color of the dog is brown.\nC. The color of the dog is gray.\nD. The color of the dog is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the little girl's shirt is pink."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_17.jpg",
"item_id": "sa_17",
"options": [
"The color of the little girl's shirt is pink.",
"The color of the little girl's shirt is white.",
"The color of the little girl's shirt is yellow.",
"The color of the little girl's shirt is black."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the little girl's shirt?\nOptions: \nA. The color of the little girl's shirt is pink.\nB. The color of the little girl's shirt is white.\nC. The color of the little girl's shirt is yellow.\nD. The color of the little girl's shirt is black.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the little girl's shirt?\nOptions: \nA. The color of the little girl's shirt is pink.\nB. The color of the little girl's shirt is white.\nC. The color of the little girl's shirt is yellow.\nD. The color of the little girl's shirt is black.\n\nOptions:\nA. The color of the little girl's shirt is pink.\nB. The color of the little girl's shirt is white.\nC. The color of the little girl's shirt is yellow.\nD. The color of the little girl's shirt is black.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the car is silver."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_4020.jpg",
"item_id": "sa_4020",
"options": [
"The color of the car is silver.",
"The color of the car is black.",
"The color of the car is blue.",
"The color of the car is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the car?\nOptions: \nA. The color of the car is silver.\nB. The color of the car is black.\nC. The color of the car is blue.\nD. The color of the car is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the car?\nOptions: \nA. The color of the car is silver.\nB. The color of the car is black.\nC. The color of the car is blue.\nD. The color of the car is red.\n\nOptions:\nA. The color of the car is silver.\nB. The color of the car is black.\nC. The color of the car is blue.\nD. The color of the car is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The pose of the woman with yellow backpack is squatting."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_33579.jpg",
"item_id": "sa_33579",
"options": [
"The pose of the woman with yellow backpack is squatting.",
"The pose of the woman with yellow backpack is standing.",
"The pose of the woman with yellow backpack is walking.",
"The pose of the woman with yellow backpack is running."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the pose of the woman with yellow backpack?\nOptions: \nA. The pose of the woman with yellow backpack is squatting.\nB. The pose of the woman with yellow backpack is standing.\nC. The pose of the woman with yellow backpack is walking.\nD. The pose of the woman with yellow backpack is running.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the pose of the woman with yellow backpack?\nOptions: \nA. The pose of the woman with yellow backpack is squatting.\nB. The pose of the woman with yellow backpack is standing.\nC. The pose of the woman with yellow backpack is walking.\nD. The pose of the woman with yellow backpack is running.\n\nOptions:\nA. The pose of the woman with yellow backpack is squatting.\nB. The pose of the woman with yellow backpack is standing.\nC. The pose of the woman with yellow backpack is walking.\nD. The pose of the woman with yellow backpack is running.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the toothbrush is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_65476.jpg",
"item_id": "sa_65476",
"options": [
"The color of the toothbrush is red.",
"The color of the toothbrush is green.",
"The color of the toothbrush is yellow.",
"The color of the toothbrush is blue."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the toothbrush?\nOptions: \nA. The color of the toothbrush is red.\nB. The color of the toothbrush is green.\nC. The color of the toothbrush is yellow.\nD. The color of the toothbrush is blue.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the toothbrush?\nOptions: \nA. The color of the toothbrush is red.\nB. The color of the toothbrush is green.\nC. The color of the toothbrush is yellow.\nD. The color of the toothbrush is blue.\n\nOptions:\nA. The color of the toothbrush is red.\nB. The color of the toothbrush is green.\nC. The color of the toothbrush is yellow.\nD. The color of the toothbrush is blue.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the suitcase is brown."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_24546.jpg",
"item_id": "sa_24546",
"options": [
"The color of the suitcase is brown.",
"The color of the suitcase is black.",
"The color of the suitcase is gray.",
"The color of the suitcase is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the suitcase?\nOptions: \nA. The color of the suitcase is brown.\nB. The color of the suitcase is black.\nC. The color of the suitcase is gray.\nD. The color of the suitcase is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the suitcase?\nOptions: \nA. The color of the suitcase is brown.\nB. The color of the suitcase is black.\nC. The color of the suitcase is gray.\nD. The color of the suitcase is red.\n\nOptions:\nA. The color of the suitcase is brown.\nB. The color of the suitcase is black.\nC. The color of the suitcase is gray.\nD. The color of the suitcase is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the tablecloth is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_58805.jpg",
"item_id": "sa_58805",
"options": [
"The color of the tablecloth is green.",
"The color of the tablecloth is white.",
"The color of the tablecloth is red.",
"The color of the tablecloth is blue."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the tablecloth?\nOptions: \nA. The color of the tablecloth is green.\nB. The color of the tablecloth is white.\nC. The color of the tablecloth is red.\nD. The color of the tablecloth is blue.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the tablecloth?\nOptions: \nA. The color of the tablecloth is green.\nB. The color of the tablecloth is white.\nC. The color of the tablecloth is red.\nD. The color of the tablecloth is blue.\n\nOptions:\nA. The color of the tablecloth is green.\nB. The color of the tablecloth is white.\nC. The color of the tablecloth is red.\nD. The color of the tablecloth is blue.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the van is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10204.jpg",
"item_id": "sa_10204",
"options": [
"The color of the van is red.",
"The color of the van is silver.",
"The color of the van is black.",
"The color of the van is white."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the van?\nOptions: \nA. The color of the van is red.\nB. The color of the van is silver.\nC. The color of the van is black.\nD. The color of the van is white.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the van?\nOptions: \nA. The color of the van is red.\nB. The color of the van is silver.\nC. The color of the van is black.\nD. The color of the van is white.\n\nOptions:\nA. The color of the van is red.\nB. The color of the van is silver.\nC. The color of the van is black.\nD. The color of the van is white.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the cleaning cloth is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_7347.jpg",
"item_id": "sa_7347",
"options": [
"The color of the cleaning cloth is red.",
"The color of the cleaning cloth is blue.",
"The color of the cleaning cloth is white.",
"The color of the cleaning cloth is green."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the cleaning cloth?\nOptions: \nA. The color of the cleaning cloth is red.\nB. The color of the cleaning cloth is blue.\nC. The color of the cleaning cloth is white.\nD. The color of the cleaning cloth is green.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the cleaning cloth?\nOptions: \nA. The color of the cleaning cloth is red.\nB. The color of the cleaning cloth is blue.\nC. The color of the cleaning cloth is white.\nD. The color of the cleaning cloth is green.\n\nOptions:\nA. The color of the cleaning cloth is red.\nB. The color of the cleaning cloth is blue.\nC. The color of the cleaning cloth is white.\nD. The color of the cleaning cloth is green.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the bus is white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_34040.jpg",
"item_id": "sa_34040",
"options": [
"The color of the bus is white.",
"The color of the bus is red.",
"The color of the bus is blue.",
"The color of the bus is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the bus?\nOptions: \nA. The color of the bus is white.\nB. The color of the bus is red.\nC. The color of the bus is blue.\nD. The color of the bus is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the bus?\nOptions: \nA. The color of the bus is white.\nB. The color of the bus is red.\nC. The color of the bus is blue.\nD. The color of the bus is yellow.\n\nOptions:\nA. The color of the bus is white.\nB. The color of the bus is red.\nC. The color of the bus is blue.\nD. The color of the bus is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the SUV car is silver."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_10919.jpg",
"item_id": "sa_10919",
"options": [
"The color of the SUV car is silver.",
"The color of the SUV car is black.",
"The color of the SUV car is red.",
"The color of the SUV car is blue."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the SUV car?\nOptions: \nA. The color of the SUV car is silver.\nB. The color of the SUV car is black.\nC. The color of the SUV car is red.\nD. The color of the SUV car is blue.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the SUV car?\nOptions: \nA. The color of the SUV car is silver.\nB. The color of the SUV car is black.\nC. The color of the SUV car is red.\nD. The color of the SUV car is blue.\n\nOptions:\nA. The color of the SUV car is silver.\nB. The color of the SUV car is black.\nC. The color of the SUV car is red.\nD. The color of the SUV car is blue.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the bottle cap is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_66471.jpg",
"item_id": "sa_66471",
"options": [
"The color of the bottle cap is red.",
"The color of the bottle cap is blue.",
"The color of the bottle cap is white.",
"The color of the bottle cap is orange."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the bottle cap?\nOptions: \nA. The color of the bottle cap is red.\nB. The color of the bottle cap is blue.\nC. The color of the bottle cap is white.\nD. The color of the bottle cap is orange.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the bottle cap?\nOptions: \nA. The color of the bottle cap is red.\nB. The color of the bottle cap is blue.\nC. The color of the bottle cap is white.\nD. The color of the bottle cap is orange.\n\nOptions:\nA. The color of the bottle cap is red.\nB. The color of the bottle cap is blue.\nC. The color of the bottle cap is white.\nD. The color of the bottle cap is orange.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the baby carriage is black."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_63876.jpg",
"item_id": "sa_63876",
"options": [
"The color of the baby carriage is black.",
"The color of the baby carriage is green.",
"The color of the baby carriage is red.",
"The color of the baby carriage is blue."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the baby carriage?\nOptions: \nA. The color of the baby carriage is black.\nB. The color of the baby carriage is green.\nC. The color of the baby carriage is red.\nD. The color of the baby carriage is blue.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the baby carriage?\nOptions: \nA. The color of the baby carriage is black.\nB. The color of the baby carriage is green.\nC. The color of the baby carriage is red.\nD. The color of the baby carriage is blue.\n\nOptions:\nA. The color of the baby carriage is black.\nB. The color of the baby carriage is green.\nC. The color of the baby carriage is red.\nD. The color of the baby carriage is blue.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the Apple logo is polychromatic."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_33492.jpg",
"item_id": "sa_33492",
"options": [
"The color of the Apple logo is polychromatic.",
"The color of the Apple logo is white.",
"The color of the Apple logo is silver.",
"The color of the Apple logo is red."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the Apple logo?\nOptions: \nA. The color of the Apple logo is polychromatic.\nB. The color of the Apple logo is white.\nC. The color of the Apple logo is silver.\nD. The color of the Apple logo is red.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the Apple logo?\nOptions: \nA. The color of the Apple logo is polychromatic.\nB. The color of the Apple logo is white.\nC. The color of the Apple logo is silver.\nD. The color of the Apple logo is red.\n\nOptions:\nA. The color of the Apple logo is polychromatic.\nB. The color of the Apple logo is white.\nC. The color of the Apple logo is silver.\nD. The color of the Apple logo is red.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the helmet is white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_42585.jpg",
"item_id": "sa_42585",
"options": [
"The color of the helmet is white.",
"The color of the helmet is black.",
"The color of the helmet is red.",
"The color of the helmet is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the helmet?\nOptions: \nA. The color of the helmet is white.\nB. The color of the helmet is black.\nC. The color of the helmet is red.\nD. The color of the helmet is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the helmet?\nOptions: \nA. The color of the helmet is white.\nB. The color of the helmet is black.\nC. The color of the helmet is red.\nD. The color of the helmet is yellow.\n\nOptions:\nA. The color of the helmet is white.\nB. The color of the helmet is black.\nC. The color of the helmet is red.\nD. The color of the helmet is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the bicycle is red."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_30179.jpg",
"item_id": "sa_30179",
"options": [
"The color of the bicycle is red.",
"The color of the bicycle is black.",
"The color of the bicycle is white.",
"The color of the bicycle is yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the bicycle?\nOptions: \nA. The color of the bicycle is red.\nB. The color of the bicycle is black.\nC. The color of the bicycle is white.\nD. The color of the bicycle is yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the bicycle?\nOptions: \nA. The color of the bicycle is red.\nB. The color of the bicycle is black.\nC. The color of the bicycle is white.\nD. The color of the bicycle is yellow.\n\nOptions:\nA. The color of the bicycle is red.\nB. The color of the bicycle is black.\nC. The color of the bicycle is white.\nD. The color of the bicycle is yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the cart is green."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_50442.jpg",
"item_id": "sa_50442",
"options": [
"The color of the cart is green.",
"The color of the cart is white.",
"The color of the cart is silver.",
"The color of the cart is black."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the cart?\nOptions: \nA. The color of the cart is green.\nB. The color of the cart is white.\nC. The color of the cart is silver.\nD. The color of the cart is black.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the cart?\nOptions: \nA. The color of the cart is green.\nB. The color of the cart is white.\nC. The color of the cart is silver.\nD. The color of the cart is black.\n\nOptions:\nA. The color of the cart is green.\nB. The color of the cart is white.\nC. The color of the cart is silver.\nD. The color of the cart is black.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the handbag is white."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_42562.jpg",
"item_id": "sa_42562",
"options": [
"The color of the handbag is white.",
"The color of the handbag is red.",
"The color of the handbag is black.",
"The color of the handbag is brown."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the handbag?\nOptions: \nA. The color of the handbag is white.\nB. The color of the handbag is red.\nC. The color of the handbag is black.\nD. The color of the handbag is brown.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the handbag?\nOptions: \nA. The color of the handbag is white.\nB. The color of the handbag is red.\nC. The color of the handbag is black.\nD. The color of the handbag is brown.\n\nOptions:\nA. The color of the handbag is white.\nB. The color of the handbag is red.\nC. The color of the handbag is black.\nD. The color of the handbag is brown.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
| |
vstar_direct_attributes
|
{
"answer_letter": "A",
"ground_truth": "A. The color of the flag is blue and yellow."
}
|
{
"bbox": [],
"image_path": "/scratch/doqihu/work/eval_data/vstar_bench/direct_attributes/sa_25219.jpg",
"item_id": "sa_25219",
"options": [
"The color of the flag is blue and yellow.",
"The color of the flag is green and white.",
"The color of the flag is blue and red.",
"The color of the flag is red and yellow."
],
"prompt_preset": "seal_zoom",
"question": "Question: What is the color of the flag?\nOptions: \nA. The color of the flag is blue and yellow.\nB. The color of the flag is green and white.\nC. The color of the flag is blue and red.\nD. The color of the flag is red and yellow.\n",
"target_object": [],
"test_type": "direct_attributes"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"},\"target_image\":{\"type\":\"integer\",\"description\":\"Specifies which image to operate on. Negative values (-1, -2) refer to the last N processed images. Positive values refer to images in sequence: 0=original image, 1=after first tool call, etc. Default is -1 (last processed image).\",\"default\":-1}},\"required\":[\"bbox_2d\"]}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**:\n<tool_call>\n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [50, 100, 200, 300], \"label\": \"the object in the center\", \"target_image\": 0}}\n</tool_call>",
"role": "system"
},
{
"content": "<image>\nQuestion: What is the color of the flag?\nOptions: \nA. The color of the flag is blue and yellow.\nB. The color of the flag is green and white.\nC. The color of the flag is blue and red.\nD. The color of the flag is red and yellow.\n\nOptions:\nA. The color of the flag is blue and yellow.\nB. The color of the flag is green and white.\nC. The color of the flag is blue and red.\nD. The color of the flag is red and yellow.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
visual_reasoning
|
point_toolbox_v6
|
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 15