Ardelyo/dara-v1

🪔 DARA — Detect & Assist Recognition AI

"Mata untuk semua" — Eyes for everyone

Lightweight Vision-Language Model for Assistive Technology

Demo • GitHub • Docs

---

⚡ What is DARA?

DARA is a lightweight VLM designed to help visually impaired users understand their surroundings through 5 specialized modes — with voice output support.

Feature	Spec
🧠 Base Model	Florence-2-base
📦 Size	232M params (~500MB)
⚡ Speed	<200ms on CPU
🌍 Languages	English, Indonesian
🔊 Output	Text + Voice (TTS)

---

🎯 5 Intelligence Modes

Mode	Use Case	Example Output
👁️ Scene	Describe surroundings	"Kitchen with wooden table. Stove on left."
😊 Emotion	Read facial expressions	"Happy. They seem in good spirits!"
💊 Medicine	Read medicine labels	"Dosage: 500mg. Take as prescribed."
💵 Currency	Identify money	"Rp 50,000. Blue-colored note."
📖 Text	OCR for signs/labels	"EXIT sign detected."

---

🚀 Quick Start

from dara import DARA

# Initialize
dara = DARA()

# Use any mode
result = dara.detect(
    image_path="photo.jpg",
    mode="scene",     # scene | emotion | medicine | currency | text
    language="en"     # en | id
)

print(result["result"])  # Text description
# Audio saved to result["audio"]

---

📥 Installation

pip install torch transformers pillow gtts
git clone https://github.com/ardelyo/dara.git

---

⚠️ Important Notes

🏥 Medical Disclaimer: Medicine mode is for reference only. Always consult healthcare professionals.

🔒 Privacy: All processing runs locally. No images are uploaded.

---

📊 Performance

Device	Latency
CPU (i7)	~180ms
GPU (RTX 3060)	~45ms
Mobile	~320ms

---

📝 Citation

@misc{dara2024,
  title={DARA: Detect & Assist Recognition AI},
  author={Ardelyo},
  year={2024},
  url={https://github.com/ardelyo/dara}
}

---

Made with ❤️ for Accessibility

⭐ GitHub • 🤗 Demo