Skip to content

CHANGELOG.md

Latest commit

 

History

History
144 lines (121 loc) · 5.26 KB

CHANGELOG.md

File metadata and controls

144 lines (121 loc) · 5.26 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.1] - 2025-01-06

Fixed

  • FFmpeg NoneType error in syncnet_pipeline.py when ffmpeg_bin is None
  • Improved error handling for FFmpeg subprocess calls
  • Fixed import paths for direct script execution (relative imports with fallback)

Changed

  • Replaced subprocess.call with subprocess.run for better error handling
  • Added explicit FFmpeg binary path validation
  • Improved error messages for FFmpeg-related failures

[0.2.0] - 2025-01-24

Added

  • Abstract base classes for all major components (BaseModel, AudioProcessor, VideoProcessor, etc.)
  • Comprehensive exception hierarchy with specific error types
  • Configuration management system with validation
  • Structured logging with color support and JSON output
  • Memory management utilities and monitoring
  • Parallel video processing support
  • Streaming audio processing capabilities
  • Optimized synchronization analyzer with caching
  • Factory pattern for model creation
  • Batch processing utilities
  • YAML/JSON configuration support
  • Detailed analysis testing with per-crop offsets and min distances
  • Comprehensive functionality verification tests

Changed

  • Complete refactoring of core modules with clean architecture
  • Separated concerns into distinct modules (audio, video, models, sync_analyzer)
  • Improved error handling throughout the codebase
  • Added input validation and sanitization
  • Optimized tensor operations for better performance
  • Modernized code with Python 3.13 features while maintaining 3.9+ compatibility
  • Fixed import paths in syncnet_python module for proper functionality
  • Updated type definitions with Python 3.9 compatibility fallbacks

Removed

  • Duplicate code in script/ directory
  • Build artifacts and temporary files
  • Redundant documentation files
  • Hard-coded configuration values
  • Intermediate test files and temporary outputs

Fixed

  • Memory leaks in video processing
  • Thread safety issues in parallel processing
  • Improper resource cleanup
  • Import path issues in syncnet_pipeline.py
  • Legacy model compatibility layer
  • PyYAML and psutil optional dependencies

Tested

  • Successfully verified all APIs with example video and audio files
  • Confirmed face detection working at 100% success rate (134 frames)
  • Validated audio-visual sync analysis with 1 frame offset detection
  • Tested command line interface functionality
  • Verified error handling and edge cases
  • Performance tested: 191.4 fps processing, 0.646s compute time

Performance Results

  • Example video: example/video.avi (134 frames)
  • Example audio: example/speech.wav
  • AV Offset detected: 1 frame (excellent sync)
  • Confidence: 4.500 (very high)
  • Min distance: 9.316
  • Face detection confidence: 0.959-1.000 range
  • Processing speed: 191.4 fps

[0.1.1] - 2025-01-24

Added

  • Proper attribution to the original SyncNet implementation by Joon Son Chung
  • Python 3.9 compatibility fixes for type hints with fallback imports
  • PYPI_RELEASE_NOTES.md for tracking PyPI releases

Changed

  • Updated project description to clarify this is an updated version for modern Python versions
  • Modified type definitions in syncnet/core/types.py to support Python 3.9+

Fixed

  • Type hint compatibility issues for Python versions prior to 3.10

[0.1.0] - 2025-01-24

Added

  • Complete refactoring for Python 3.13 compatibility
  • Modern Python features:
    • Type hints throughout the codebase
    • Async/await support for concurrent processing
    • Exception groups for better error handling
    • Dataclasses for configuration management
  • Modular package structure:
    • syncnet/core/ - Refactored core components
    • syncnet_python/ - Legacy compatibility wrapper
  • PyPI package configuration with pyproject.toml
  • Comprehensive documentation:
    • README.md with installation and usage instructions
    • CLAUDE.md for AI-assisted development
  • Development tools configuration:
    • Black formatter settings
    • Ruff linter configuration
    • MyPy type checking setup

Changed

  • Migrated from setup.py to modern pyproject.toml configuration
  • Replaced deprecated torch.autograd.Variable with direct tensor usage
  • Updated S3FD face detector model loading with proper state dict mapping
  • Reorganized directory structure for better maintainability
  • Cleaned up intermediate files and temporary outputs

Fixed

  • PyTorch deprecation warnings
  • S3FD model state dict loading issues with parameter name mapping
  • Legacy model compatibility through dedicated compatibility layer

Tested

  • Successfully tested on example video:
    • AV offset: 1 frame
    • Confidence: 4.568
  • Verified functionality matches original implementation

Technical Details

  • Python 3.13 primary development target
  • Backward compatibility to Python 3.9
  • Dependencies:
    • PyTorch >= 2.0.0
    • OpenCV >= 4.8.0
    • NumPy >= 1.24.0
    • SciPy >= 1.10.0
    • Other supporting libraries

Notes

  • Model weights (sfd_face.pth, syncnet_v2.model) must be downloaded separately due to PyPI size limitations
  • Primary testing done on Python 3.13; other versions may have minor compatibility issues