# VideoBackgroundReplacer2 Deployment Guide

This guide provides instructions for deploying the VideoBackgroundReplacer2 application to Hugging Face Spaces with GPU acceleration.

## Prerequisites

- Docker
- Git
- Python 3.8+
- NVIDIA Container Toolkit (for local GPU testing)
- Hugging Face account with access to GPU Spaces

## Local Development

### 1. Clone the repository
```bash
git clone <repository-url>
cd VideoBackgroundReplacer2
```

### 2. Build the Docker image
```bash
# Make the build script executable
chmod +x build_and_deploy.sh

# Build the image
./build_and_deploy.sh
```

### 3. Run the container locally
```bash
docker run --gpus all -p 7860:7860 -v $(pwd)/checkpoints:/home/user/app/checkpoints videobackgroundreplacer2:latest
```

## Hugging Face Spaces Deployment

### 1. Create a new Space
- Go to [Hugging Face Spaces](https://huggingface.co/spaces)
- Click "Create new Space"
- Select "Docker" as the SDK
- Choose a name and set the space to private if needed
- Select GPU as the hardware

### 2. Configure the Space
Add the following environment variables to your Space settings:
- `SAM2_DEVICE`: `cuda`
- `MATANY_DEVICE`: `cuda`
- `PYTORCH_CUDA_ALLOC_CONF`: `max_split_size_mb:256,garbage_collection_threshold:0.8`
- `TORCH_CUDA_ARCH_LIST`: `7.5 8.0 8.6+PTX`

### 3. Deploy to Hugging Face
```bash
# Set your Hugging Face token
export HF_TOKEN=your_hf_token
export HF_USERNAME=your_username

# Build and deploy
./build_and_deploy.sh
```

## Health Check

You can verify the installation by running:
```bash
docker run --rm videobackgroundreplacer2:latest python3 health_check.py
```

## Troubleshooting

### Build Failures
- Ensure you have enough disk space (at least 10GB free)
- Check Docker logs for specific error messages
- Verify your internet connection is stable

### Runtime Issues
- Check container logs: `docker logs <container_id>`
- Verify GPU is detected: `nvidia-smi` inside the container
- Check disk space: `df -h`

## Performance Optimization

- For faster inference, use the `sam2_hiera_tiny` model
- Adjust batch size based on available GPU memory
- Enable gradient checkpointing for large models

## Monitoring

- Use `nvidia-smi` to monitor GPU usage
- Check container logs for any warnings or errors
- Monitor memory usage with `htop` or similar tools