| # ISL command recognition service (FastAPI + browser dashboard + CLI tools). | |
| # | |
| # What works inside a container: dataset audit, training, evaluation, robustness, | |
| # the FastAPI service, browser dashboard, and batch inference over video files. | |
| # | |
| # What does NOT work inside a container by default: a *local* webcam | |
| # (`backend/inference/realtime.py`) - containers usually have no /dev/video0. | |
| # The browser dashboard uses browser camera access and sends sampled frames to | |
| # the API. See README's webcam limitations for the supported local workflow. | |
| FROM python:3.11-slim | |
| ENV PYTHONDONTWRITEBYTECODE=1 \ | |
| PYTHONUNBUFFERED=1 \ | |
| PIP_NO_CACHE_DIR=1 \ | |
| TF_CPP_MIN_LOG_LEVEL=2 \ | |
| OPENCV_VIDEOIO_PRIORITY_LIST=FFMPEG \ | |
| ISL_PERSIST_UPLOADS=false | |
| # OpenCV headless needs these shared libraries; ffmpeg decodes uploaded videos. | |
| RUN apt-get update && apt-get install -y --no-install-recommends \ | |
| ffmpeg libegl1 libgl1 libglib2.0-0 libsm6 libxext6 libxrender1 \ | |
| curl ca-certificates \ | |
| && rm -rf /var/lib/apt/lists/* | |
| WORKDIR /app | |
| COPY requirements.txt ./ | |
| RUN pip install --upgrade pip && pip install -r requirements.txt | |
| COPY . . | |
| # The project includes the model weights and MediaPipe task bundles for the demo. | |
| # Fetch bundles only if a task asset is missing. | |
| RUN python scripts/download_models.py --check || python scripts/download_models.py | |
| # Container user UID 1000 for Hugging Face Spaces / general container runtimes. | |
| RUN useradd --create-home --uid 1000 isl && chown -R isl:isl /app | |
| USER isl | |
| EXPOSE 7860 | |
| HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \ | |
| CMD curl -fsS http://127.0.0.1:${PORT:-7860}/health || exit 1 | |
| # The FastAPI service serves the browser dashboard on $PORT (defaults to 7860 for Hugging Face Spaces). | |
| CMD ["sh", "-c", "uvicorn backend.main:app --host 0.0.0.0 --port ${PORT:-7860}"] | |