camino-gr/blip2-visual-qa-demo

BLIP-2 Visual QA Demo

A visual question answering demo powered by Salesforce/blip2-opt-2.7b.

🚀 Live Demo

👉 [Try it on Hugging Face Spaces](https://huggingface.co/spaces/camino-gr/blip2-visual-qa-demo)

Features

1. Image Captioning

Upload an image and get an automatic caption describing its contents.

2. Visual QA

Upload an image and ask a question about it — the model will answer based on visual understanding.

3. Image Comparison

Upload two images and get descriptions of each, along with highlighted differences.

Model

This demo uses BLIP-2 (Bootstrapping Language-Image Pre-training) with the OPT-2.7B language model backbone. BLIP-2 bridges the modality gap between vision and language using a lightweight Querying Transformer (Q-Former).

Tech Stack

• —Model: Salesforce/blip2-opt-2.7b via 🤗 Transformers
• —UI: Gradio 5
• —Runtime: Docker (Python 3.12)

Local Development

pip install -r requirements.txt
python app.py

License

MIT