Skip to content

v0.2.5 - Initial public release

Latest
Latest

Choose a tag to compare

View all tags
@YunyaoYan YunyaoYan released this 02 Jun 03:07
· 5 commits to main since this release
0.2.5
b03fe0a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

codec-video-prep v0.2.5

Initial public release of codec-video-prep, a codec-aware video preprocessing package for LLaVA-OneVision style video inputs.

Highlights

  • Added patched FFmpeg support for extracting H.264 / HEVC / VP9 bitcost maps.
  • Added cv_reader_fast, a native extension that can fetch decoded BGR frames and codec bitcost data in the same decode scan.
  • Added the codec-video-prep CLI for generating patch canvases and metadata from input videos.
  • Added readiness-based frame grouping and top-k patch canvas packing.
  • Added Linux manylinux wheel build support for x86_64 and aarch64.
  • Added codec-video-prep-doctor for installation diagnostics.

Notes

  • The recommended decode backend is cv_reader_pixels.
  • Source builds should use build_pixel_ffmpeg.sh.
  • macOS wheels are not provided in this release.
Assets 10
Loading