README.md

metadata

title: T2AV Eval UI
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860

Video Evaluation UI

This app is configured to deploy as a Hugging Face Docker Space. The Docker build compiles the Vite frontend, serves it from FastAPI, and stores each annotation as a JSON file in a Hugging Face Dataset repo.

Hugging Face Space Deployment

1. Create a new Space on Hugging Face and choose Docker as the SDK.
2. Add a Space secret named HF_TOKEN with a Hugging Face token that has write access to datasets in the dghadiya namespace (the token must belong to, or be added as a collaborator on, that account — see note below).
3. Optional: set HF_ANNOTATIONS_DATASET_REPO if you want to use a dataset repo other than dghadiya/t2av_eval_annotations. Optional: set HF_VIDEO_DATASET_REPO if the evaluation videos move to a different dataset repo (default dghadiya/T2AV_TemporalConstraint; this one is public, so no token is required just to read it).
4. Add this repository as the Space repository or push it to the Space remote:

git remote add hf https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME
git push hf main

5. Hugging Face will build the root Dockerfile and expose the app on port 7860.

The backend creates the annotation dataset repo on first save if it does not already exist. By default it creates the dataset as private. Set HF_ANNOTATIONS_DATASET_PRIVATE=false to create a public dataset instead.

Note: HF_TOKEN must belong to (or have write access on) whatever account owns HF_ANNOTATIONS_DATASET_REPO — a token cannot create or write repos in a namespace its owner doesn't control. To check which account a token belongs to: python -c "from huggingface_hub import HfApi; print(HfApi().whoami(token='hf_...'))".

Backend

cd backend

pip install -r requirements.txt

uvicorn main:app --reload

Backend runs at: http://localhost:8000

Health check: http://localhost:8000/health

Frontend

cd frontend

npm install

npm run dev

Frontend runs at: http://localhost:5173