MGRS-200k Dataset

MGRS-200k is the first multi-granularity remote sensing (RS) image-text dataset, introduced in the paper FarSLIP: Discovering Effective CLIP Adaptation for Fine-Grained Remote Sensing Understanding. It features rich object-level textual supervision for RS region-category alignment. This dataset was constructed to address limitations in existing RS image-text datasets, particularly their limited spatial awareness.

Each image in MGRS-200k is annotated with both short and long global-level captions, along with multiple object-category pairs. The dataset is instrumental for fine-grained vision-language alignment in the RS domain and supports a range of tasks including open-vocabulary semantic segmentation, zero-shot classification, and image-text retrieval.

Paper and Code

• Paper: FarSLIP: Discovering Effective CLIP Adaptation for Fine-Grained Remote Sensing Understanding
• Code: https://github.com/NJU-LHRS/FarSLIP

Dataset Structure

The MGRS-200k dataset is composed of remote sensing images and corresponding textual information. Based on the training scripts, the typical structure includes:

• global_imgs/: A directory containing the image files.
• text_info.json: A JSON file holding the textual annotations, which include global captions and object-category pairs.

An illustrative example of the dataset's multi-granularity annotations:

Examples from MGRS-200k

Sample Usage

MGRS-200k is designed for training and fine-tuning vision-language models for remote sensing tasks. Below is an example command from the associated GitHub repository, demonstrating how MGRS-200k is used in the FarSLIP training pipeline (Stage 2):

torchrun --nproc_per_node=4 -m open_clip_train.main \
--train-dataset-name MGRS \
--root-train-img-dir '/your/path/to/mgrs/global_imgs/' \
--train-data '/your/path/to/mgrs/text_info.json' \
--train-dataset-type json \
--method farslip2 \
--warmup 250 \
--batch-size 40 \
--lr 4e-9 \
--wd 1.0 \
--epochs 10 \
--model ViT-B-16 \
--loss-type global_itc local_itc \
--local-itc-align cls

This command specifies:

• --train-dataset-name MGRS: Identifies the dataset being used.
• --root-train-img-dir: Path to the directory containing the image files.
• --train-data: Path to the JSON file with textual annotations.
• --train-dataset-type json: Specifies that dataset metadata is loaded from a JSON file.

Citation

If you find this dataset or the associated work useful, please consider citing the paper:

@article{li2025farslip,
title={FarSLIP: Discovering Effective CLIP Adaptation for Fine-Grained Remote Sensing Understanding},
author={Zhenshi Li and Weikang Yu and Dilxat Muhtar and Xueliang Zhang and Pengfeng Xiao and Pedram Ghamisi and Xiao Xiang Zhu},
journal={arXiv preprint arXiv:2511.14901},
year={2025}
}