A Pure-Python Computer Vision Library That's Fast and Minimal

What's Changed (Highlights)

📓 Release Notes can be found here ↗️

New Features ✨

• ✨ FFGear API:
  • Introduced a brand-new high-performance, multi-threaded video decoding Gear built on top of DeFFcode's FFdecoder API.
  • Supports real-time FFmpeg subprocess pipelines, low-overhead hardware-accelerated decoding (e.g., CUDA/NVDEC), flexible pixel format conversions, and asynchronous per-frame metadata extraction.
  • Includes dedicated parameters: source_demuxer, frame_format, custom_ffmpeg, and extract_metadata.
• ⚡️ Unified Backend Selection (Backend Enum):
  • Introduced a unified api parameter accepting a new Backend enum (Backend.CAMGEAR, Backend.PIGEAR, Backend.FFGEAR) across VideoGear, WebGear, WebGear_RTC, and NetGear_Async.
  • Replaces the legacy enablePiCamera boolean flag with a clean, explicit enumeration-based backend selection.
• ⚡️ Plugin-Ready Stabilizer Architecture:
  • Refactored the monolithic Stabilizer class into a plugin-style stabilizer/ sub-package with a shared _StabilizerBase, an ASWStabilizer implementation, and a factory Stabilizer() function.
  • Added a new StabilizerMode enum for selecting stabilization backends via VideoGear's STABILIZER_MODE option, with clean provisions for upcoming real-time backends (e.g., Kalman tracking).
• 🚩 WriteGear: Discard -vcodec via None:
  • Setting -vcodec or -c:v to None in output_params now cleanly omits the video codec parameter, letting FFmpeg auto-select the encoder (useful for GIF output and similar workflows).
• 📦️ Poetry Package Management:
  • Added an official Poetry installation guide and ecosystem support for Poetry-managed environments.

Updates/Improvements ⚡️

• Packaging Infrastructure Modernization:
  • Migrated build and metadata configurations from legacy setup.py/setup.cfg to modern pyproject.toml configuration standards.
  • Switched to automatic package discovery via setuptools, excluding tests and docs from distribution builds.
• Toolchain Overhaul (Ruff Migration):
  • Completely replaced flake8 and black with ruff for unified, high-speed linting and formatting across the entire codebase.
• Type Hint Modernization:
  • Systematically updated type hints to leverage native Python 3.10+ structures, including PEP 604 union types (X | Y) and built-in collection generics.
• PEP 561 Compliance:
  • Added a py.typed marker file so external type checkers (e.g., pyright, mypy) properly recognize VidGear as an inline-typed package.
• CamGear YouTube Backend Refactor:
  • Extracted YT_backend into a dedicated yt_backend module for cleaner separation of concerns and improved maintainability.
• Asyncio CLI Updates:
  • Replaced boolean string arguments with proper store_true flags for -S, -yt, -l, and -ep.
  • Added -a/--api flag for backend selection (camgear/pigear/ffgear) and FFGear-specific arguments (-sd, -ff, -cf).
  • Marked --enablePiCamera as a deprecated CLI flag with a runtime warning.

Breaking Updates/Changes 💥

• 💣 Deprecated enablePiCamera:

  • The enablePiCamera boolean flag has been officially deprecated across all core capture APIs (VideoGear, WebGear, WebGear_RTC, NetGear_Async) in favor of the explicit api=Backend.PIGEAR enumeration syntax.
  • Using the old flag will emit a DeprecationWarning at runtime. It will be removed in a future release.

Bug-fixes 🐛

• Stabilizer Memory Leak:
  • Fixed a severe memory bottleneck where self.__transforms tracking records grew unbounded over the entire runtime duration (O(total_frames) → O(smoothing_radius)), eliminating an out-of-memory hazard for long-running sessions.
• FFGear Metadata Tuple Handling:
  • Fixed read() to correctly enqueue and return (frame, metadata) tuples when extract_metadata mode is enabled.
• Helper Utilities:
  • Fixed validate_video to verify file path existence before processing.
  • Fixed missing demuxer header handling in get_supported_demuxers.
  • Corrected StopIteration handling in default video codec selection logic.
  • Fixed IndexError in validate_audio bit-depth parsing by replacing bare re.findall indexing with guarded re.search and adding a fallback parser for planar sample formats (fltp, s16p, u8p, etc.).
  • Fixed get_supported_demuxers to expand all comma-separated demuxer aliases (e.g. "matroska,webm" now yields both entries instead of only the last).

VidGear is free and open source and will always remain so. ❤️

It is something I am doing with my own free time. But so much more needs to be done, and I need your help to do this. For just the price of a cup of coffee 🍵, you can make a difference 🤝

Detailed Changelog: https://github.com/abhiTronix/vidgear/compare/vidgear-0.3.4...vidgear-0.3.5