Dockerfile

# Base image with a lightweight version suitable for Hugging Face Spaces
FROM python:3.8-slim

# Install system dependencies
RUN apt-get update && apt-get install -y \
    git \
    wget \
    && rm -rf /var/lib/apt/lists/*

# Install system-level dependencies for OpenCV
RUN apt-get update && apt-get install -y \
    libglib2.0-0 \
    libsm6 \
    libxext6 \
    libxrender-dev \
    libgl1-mesa-glx \
    && rm -rf /var/lib/apt/lists/*

# Install Python packages including Hugging Face Transformers, TorchScript, Flask, TensorFlow, TensorFlow Hub, and OpenCV
RUN pip install --no-cache-dir \
    torch \
    torchvision \
    transformers \
    requests \
    Flask \
    Pillow \
    huggingface_hub \
    tensorflow \
    tensorflow_hub \
    opencv-python

# Set Hugging Face cache to a guaranteed writable directory
ENV TRANSFORMERS_CACHE=/tmp/cache
RUN mkdir -p /tmp/cache && chmod -R 777 /tmp/cache

# Create directories for the models
RUN mkdir -p /models/blip /models/clip

# Python script to download models using tensorflow_hub and huggingface_hub
RUN echo "import os\n\
import tensorflow_hub as hub\n\n\
# Download MoveNet model from TensorFlow Hub (loaded directly in app, not saved)\n\
movenet_model = hub.load('https://tfhub.dev/google/movenet/singlepose/lightning/4')\n\n\
# Download BLIP model and tokenizer using huggingface_hub\n\
from transformers import BlipForConditionalGeneration, BlipProcessor\n\
BlipForConditionalGeneration.from_pretrained('Salesforce/blip-image-captioning-large').save_pretrained('/models/blip')\n\
BlipProcessor.from_pretrained('Salesforce/blip-image-captioning-large').save_pretrained('/models/blip')\n\n\
# Download CLIP model and processor using huggingface_hub\n\
from transformers import CLIPModel, CLIPProcessor\n\
CLIPModel.from_pretrained('openai/clip-vit-large-patch14').save_pretrained('/models/clip')\n\
CLIPProcessor.from_pretrained('openai/clip-vit-large-patch14').save_pretrained('/models/clip')" > download_models.py

# Run the script to download models
RUN python download_models.py

# Copy the inference script (app.py) into the container
COPY app.py /app/app.py

# Expose the default port for Flask
EXPOSE 7860

# Run the Flask app
CMD ["python", "/app/app.py"]