fest

DEPLOYMENT.md

@@ -0,0 +1,90 @@
+# 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

Dockerfile

@@ -1,10 +1,23 @@
 # ===============================
 # BackgroundFX Pro — Dockerfile (CUDA 12.1.1 aligned)
+# Pinned to PyTorch 2.3.1 + CUDA 12.1
 # ===============================
 
 # ===== CHAPTER 1: Base image & user =====
 FROM nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04
 
+# Environment variables
+ENV DEBIAN_FRONTEND=noninteractive \
+    PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_DEFAULT_TIMEOUT=100 \
+    TORCH_CUDA_ARCH_LIST="7.5 8.0 8.6+PTX" \
+    TORCH_NVCC_FLAGS="-Xfatbin -compress-all" \
+    FORCE_CUDA="1" \
+    CUDA_VISIBLE_DEVICES="0"
+
 ARG SAM2_SHA=__PIN_ME__
 ARG SAM2_MODEL_ID=facebook/sam2
 ARG SAM2_VARIANT=sam2_hiera_large
@@ -25,29 +38,32 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
     git ffmpeg python3 python3-pip python3-venv \
     wget curl ca-certificates \
     libgl1-mesa-glx libglib2.0-0 libsm6 libxext6 libxrender-dev libgomp1 \
- && rm -rf /var/lib/apt/lists/*
+    libgl1 \
+    && rm -rf /var/lib/apt/lists/* \
+    && apt-get clean
 
-# ===== CHAPTER 3: Python & Torch (match CUDA 12.1.1) =====
+# ===== CHAPTER 3: Python & Core Dependencies =====
 RUN python3 -m pip install --no-cache-dir --upgrade pip setuptools wheel
-# Torch cu121 wheels (MATCH base image CUDA 12.1.1)
+
+# Install PyTorch with CUDA 12.1
 RUN python3 -m pip install --no-cache-dir \
-      --extra-index-url https://download.pytorch.org/whl/cu121 \
-      torch==2.3.1+cu121 torchvision==0.18.1+cu121 torchaudio==2.3.1+cu121 && \
-    python3 - <<'PY'
-import torch
-print("VERIFY Torch:", torch.__version__, "CUDA:", torch.version.cuda)
-assert torch.__version__.startswith("2.3.1") and torch.version.cuda.startswith("12.1")
-PY
-
-# ===== CHAPTER 4: App deps + MatAnyone =====
-COPY --chown=user requirements.txt ./requirements.txt
-# (requirements.txt intentionally has NO torch/torchvision/torchaudio)
-RUN python3 -m pip install --no-cache-dir -r requirements.txt
+    --extra-index-url https://download.pytorch.org/whl/cu121 \
+    torch==2.3.1+cu121 \
+    torchvision==0.18.1+cu121 \
+    torchaudio==2.3.1+cu121 \
+    && python3 -c "import torch; print(f'PyTorch version: {torch.__version__}')\
+    print(f'CUDA available: {torch.cuda.is_available()}')\
+    print(f'CUDA version: {torch.version.cuda if torch.cuda.is_available() else "N/A"}')\
+    print(f'cuDNN version: {torch.backends.cudnn.version() if torch.cuda.is_available() else "N/A"}')"
+
+# ===== CHAPTER 4: Install Pinned Dependencies =====
+COPY --chown=user requirements-pinned.txt ./requirements-pinned.txt
+RUN python3 -m pip install --no-cache-dir -r requirements-pinned.txt
 
 # Install MatAnyone from GitHub (following official documentation)
 RUN echo "Installing MatAnyone from GitHub (main branch)..." && \
     python3 -m pip install --no-cache-dir -v \
-    git+https://github.com/pq-yang/MatAnyone@main && \
+    git+https://github.com/pq-yang/MatAnyone@main#egg=matanyone && \
     echo "MatAnyone installation complete" && \
     python3 -c "import matanyone; print('MatAnyone imported successfully')"
 
@@ -55,7 +71,19 @@ RUN echo "Installing MatAnyone from GitHub (main branch)..." && \
 USER user
 COPY --chown=user . /home/user/app
 
-# ===== CHAPTER 6: Third-party — SAM2 (editable) =====
+# ===== CHAPTER 6: Health check and final cleanup =====
+# Copy and make health check script executable
+COPY --chown=user health_check.py /home/user/app/
+RUN chmod +x /home/user/app/health_check.py
+
+# Run health check during build
+RUN python3 /home/user/app/health_check.py
+
+# Cleanup
+RUN apt-get clean && \
+    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* /root/.cache /home/user/.cache/pip
+
+# ===== CHAPTER 7: Third-party — SAM2 (editable) =====
 USER root
 RUN rm -rf third_party/sam2 && \
     git clone --depth=1 https://github.com/facebookresearch/segment-anything-2.git third_party/sam2 && \
@@ -98,6 +126,20 @@ ENV PYTHONUNBUFFERED=1 \
     ENABLE_MATANY=1 \
     SAM2_DEVICE=cuda \
     MATANY_DEVICE=cuda \
+    # Disable unnecessary backends to reduce memory usage \
+    KMP_DETERMINISTIC_REDUCTION=1 \
+    KMP_INIT_AT_FORK=FALSE \
+    # Optimize PyTorch for inference \
+    TORCH_COMPILE_DEBUG=0 \
+    TORCHINDUCTOR_MAX_AUTOTUNE=1 \
+    # Disable debug logs \
+    TF_CPP_MIN_LOG_LEVEL=2 \
+    # Optimize OpenCV \
+    OPENCV_OPENCL_RUNTIME= \
+    OPENCV_OPENCL_DEVICE=:GPU: \
+    # Optimize NumPy \
+    NPY_NUM_BUFSIZE=8192 \
+    NPY_NUM_THREADS=2 \
     MATANY_FORCE_VIDEO=0 \
     MATANY_PREFER_VIDEO=0 \
     MATANY_FORCE_STEP=0 \
@@ -107,5 +149,5 @@ USER user
 EXPOSE 7860
 
 # ===== CHAPTER 8: Healthcheck & CMD =====
-HEALTHCHECK --interval=30s --timeout=5s --retries=5 CMD wget -qO- "http://127.0.0.1:${PORT:-7860}/" || exit 1
+HEALTHCHECK --interval=30s --timeout=5s --retries=5 CMD python3 /home/user/app/health_check.py && wget -qO- "http://127.0.0.1:${PORT:-7860}/" || exit 1
 CMD ["python3","-u","app.py"]

build_and_deploy.sh

@@ -0,0 +1,46 @@
+#!/bin/bash
+set -e
+
+# Configuration
+IMAGE_NAME="videobackgroundreplacer2"
+TAG="latest"
+REPO="huggingface/space"  # Update with your HF space repo
+
+# Colors for output
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+NC='\033[0m' # No Color
+
+echo -e "${YELLOW}=== Building VideoBackgroundReplacer2 Docker Image ===${NC}"
+
+# Build the Docker image
+echo -e "${GREEN}Building Docker image...${NC}"
+docker build -t "${IMAGE_NAME}:${TAG}" .
+
+# Verify the image was built
+if [ $? -ne 0 ]; then
+    echo -e "${YELLOW}Docker build failed. Please check the build logs.${NC}"
+    exit 1
+fi
+
+echo -e "${GREEN}Docker image built successfully!${NC}"
+
+# Push to Hugging Face Space (if configured)
+if [ -n "${HF_TOKEN}" ] && [ -n "${REPO}" ]; then
+    echo -e "${GREEN}Logging into Hugging Face Hub...${NC}"
+    echo "${HF_TOKEN}" | docker login -u "${HF_USERNAME:-user}" --password-stdin "registry.hf.space"
+    
+    echo -e "${GREEN}Tagging and pushing image to Hugging Face Space...${NC}"
+    docker tag "${IMAGE_NAME}:${TAG}" "registry.hf.space/${REPO}/app:${TAG}"
+    docker push "registry.hf.space/${REPO}/app:${TAG}"
+    
+    echo -e "${GREEN}Image pushed to Hugging Face Space!${NC}"
+else
+    echo -e "${YELLOW}HF_TOKEN or REPO not set. Skipping push to Hugging Face Space.${NC}"
+    echo "To push to Hugging Face Space, set the following environment variables:"
+    echo "  export HF_TOKEN=your_hf_token"
+    echo "  export HF_USERNAME=your_hf_username"
+    echo "  # Then run this script again"
+fi
+
+echo -e "${GREEN}=== Build and Deployment Complete ===${NC}"

health_check.py

@@ -0,0 +1,82 @@
+#!/usr/bin/env python3
+"""
+Health check script for VideoBackgroundReplacer2
+Verifies all dependencies, CUDA, and model loading
+"""
+import sys
+import torch
+import importlib.metadata
+import subprocess
+import platform
+import json
+from pathlib import Path
+
+def print_header(title):
+    print(f"\n{'='*80}\n{title}\n{'='*80}")
+
+def check_python():
+    print_header("PYTHON ENVIRONMENT")
+    print(f"Python: {sys.version}")
+    print(f"Platform: {platform.platform()}")
+    print(f"Working Directory: {Path.cwd()}")
+
+def check_cuda():
+    print_header("CUDA & GPU STATUS")
+    print(f"PyTorch version: {torch.__version__}")
+    print(f"CUDA available: {torch.cuda.is_available()}")
+    
+    if torch.cuda.is_available():
+        print(f"CUDA version: {torch.version.cuda}")
+        print(f"cuDNN version: {torch.backends.cudnn.version() if hasattr(torch.backends.cudnn, 'version') else 'N/A'}")
+        print(f"Number of GPUs: {torch.cuda.device_count()}")
+        
+        for i in range(torch.cuda.device_count()):
+            print(f"\nGPU {i}:")
+            print(f"  Name: {torch.cuda.get_device_name(i)}")
+            print(f"  Memory: {torch.cuda.get_device_properties(i).total_memory/1024**3:.2f} GB")
+    else:
+        print("WARNING: CUDA is not available. Running in CPU mode.")
+
+def check_dependencies():
+    print_header("INSTALLED PACKAGES")
+    packages = [
+        'torch', 'torchvision', 'torchaudio',
+        'numpy', 'opencv-python', 'gradio',
+        'mediapipe', 'huggingface-hub', 'matanyone',
+        'segment-anything-2'
+    ]
+    
+    for pkg in packages:
+        try:
+            version = importlib.metadata.version(pkg)
+            print(f"{pkg}: {version}")
+        except importlib.metadata.PackageNotFoundError:
+            print(f"{pkg}: NOT INSTALLED")
+
+def check_models():
+    print_header("MODEL CHECK")
+    # Add model loading checks here
+    pass
+
+def check_environment():
+    print_header("ENVIRONMENT VARIABLES")
+    env_vars = [
+        'CUDA_VISIBLE_DEVICES', 'TORCH_CUDA_ARCH_LIST',
+        'PYTORCH_CUDA_ALLOC_CONF', 'PYTHONPATH',
+        'SAM2_DEVICE', 'MATANY_DEVICE'
+    ]
+    
+    for var in env_vars:
+        print(f"{var}: {os.environ.get(var, 'Not set')}")
+
+if __name__ == "__main__":
+    try:
+        check_python()
+        check_cuda()
+        check_dependencies()
+        check_environment()
+        check_models()
+        print("\n✅ Health check completed successfully!")
+    except Exception as e:
+        print(f"\n❌ Health check failed: {str(e)}", file=sys.stderr)
+        sys.exit(1)

requirements-pinned.txt

@@ -0,0 +1,51 @@
+# Pinned requirements for Hugging Face Space deployment
+# Generated 2025-09-17
+
+# Core
+numpy==1.24.4
+Pillow==10.4.0
+protobuf==4.25.3
+
+# Image/Video Processing
+opencv-python-headless==4.8.1.78
+imageio==2.25.0
+imageio-ffmpeg==0.4.9
+moviepy==1.0.3
+decord==0.6.0
+scikit-image==0.20.0
+
+# MediaPipe
+mediapipe==0.10.21
+
+# SAM2 Dependencies
+--extra-index-url https://download.pytorch.org/whl/cu121
+git+https://github.com/facebookresearch/segment-anything-2@main#egg=segment-anything-2
+hydra-core==1.3.2
+omegaconf==2.3.0
+einops==0.8.1
+timm==1.0.19
+pyyaml==6.0.2
+matplotlib==3.10.6
+
+# MatAnyone Dependencies
+git+https://github.com/pq-yang/MatAnyone@main#egg=matanyone
+kornia==0.7.4
+tqdm==4.67.1
+
+# UI and API
+gradio==5.46.0
+
+# Helpers and Utilities
+huggingface-hub==0.35.0
+ffmpeg-python==0.2.0
+psutil==6.1.1
+requests==2.32.5
+scikit-learn==1.7.2
+
+# Additional Dependencies
+gputil==1.4.0
+nvidia-ml-py3==7.352.0
+loguru==0.7.3
+python-multipart==0.0.20
+uvicorn==0.35.0
+fastapi==0.116.2

requirements.txt

@@ -1,7 +1,8 @@
-# ===== Core Dependencies with CUDA 12.1 =====
-torch==2.3.1+cu121
-torchvision==0.18.1+cu121
-torchaudio==2.3.1+cu121
+# ===== Core Dependencies =====
+# PyTorch is installed in Dockerfile with CUDA 12.1
+# torch==2.3.1+cu121
+# torchvision==0.18.1+cu121
+# torchaudio==2.3.1+cu121
 
 # ===== Base Dependencies =====
 numpy>=1.24.0,<2.0.0
@@ -9,7 +10,7 @@ Pillow>=10.0.0,<11.0.0
 protobuf>=4.25.0,<5.0.0
 
 # ===== Image/Video Processing =====
-opencv-python-headless>=4.8.0,<5.0.0
+opencv-python-headless==4.8.1.78
 imageio>=2.25.0,<3.0.0
 imageio-ffmpeg>=0.4.7,<0.6.0
 moviepy>=1.0.3,<2.0.0
@@ -17,10 +18,10 @@ decord>=0.6.0,<0.7.0
 scikit-image>=0.19.3,<0.21.0
 
 # ===== MediaPipe =====
-mediapipe>=0.10.0,<0.11.0
+mediapipe==0.10.21
 
 # ===== SAM2 Dependencies =====
-git+https://github.com/facebookresearch/segment-anything-2@main
+git+https://github.com/facebookresearch/segment-anything-2@main#egg=segment-anything-2
 hydra-core>=1.3.2,<2.0.0
 omegaconf>=2.3.0,<3.0.0
 einops>=0.8.0,<0.9.0
@@ -29,15 +30,15 @@ pyyaml>=6.0.2,<7.0.0
 matplotlib>=3.9.2,<4.0.0
 
 # ===== MatAnyone Dependencies =====
-git+https://github.com/pq-yang/MatAnyone@main
+git+https://github.com/pq-yang/MatAnyone@main#egg=matanyone
 kornia>=0.7.3,<0.8.0
 tqdm>=4.66.5,<5.0.0
 
 # ===== UI and API =====
-gradio>=5.42.0,<6.0.0
+gradio==5.46.0
 
 # ===== Helpers and Utilities =====
-huggingface-hub>=0.33.5,<1.0.0
+huggingface-hub==0.35.0
 ffmpeg-python>=0.2.0,<1.0.0
 psutil>=6.0.0,<7.0.0
 requests>=2.32.3,<3.0.0