siddhish2006/jewelry-annotator

Jewelry Annotator

Admin tab uploads jewelry photos -> YOLO detects and crops each piece -> BLIP captions each crop -> everything lands in a searchable catalog and an Excel file. Customer tab searches that catalog by plain-language description, lets a customer choose a design, and can generate an AI variation of it with Stable Diffusion.

Running locally

See start_server.bat (Windows) - it points storage at a drive with room and starts the server. Or directly:

cd backend
pip install -r requirements.txt
uvicorn main:app --reload

Deployed on Hugging Face Spaces

This repo deploys as-is via the Dockerfile. On the free CPU tier there is no GPU, so captioning and AI image generation are slower than on the local RTX 3050 setup, and there's no persistent disk by default: uploaded photos, the database and model cache are wiped whenever the Space restarts or goes to sleep from inactivity. Add Spaces' paid persistent storage and set DATA_ROOT to its mount path to keep data across restarts.