2 Releases published by 2 people

• v0.5.0 TorchCodec 0.5

  published Jul 23, 2025

• v0.6.0 TorchCodec 0.6.0

  published Aug 7, 2025

25 Pull requests merged by 5 people

• Release 0.6

  #808 merged Aug 7, 2025

• Build Windows CPU wheel

  #753 merged Aug 5, 2025

• Remove av_q2d, remove test rounding

  #797 merged Jul 31, 2025

• Improve pytest default options

  #794 merged Jul 31, 2025

• Add support for CUDA >= 12.9

  #757 merged Jul 30, 2025

• Add 0.5 to README installation table

  #793 merged Jul 30, 2025

• Support CPU fallback for videos that don't get decoded by nvdec

  #792 merged Jul 30, 2025

• Convert negative frame indices in C++, convert slices in Python

  #746 merged Jul 29, 2025

• Consistently use ptsToSeconds

  #784 merged Jul 29, 2025

• Speed-up CMake builds by allowing a persistent Cmake build dir

  #788 merged Jul 28, 2025

• Use new way of specify the license

  #789 merged Jul 28, 2025

• Avoid silently wrong results on 10bit CUDA

  #777 merged Jul 28, 2025

• Avoid copies of frame mappings

  #785 merged Jul 24, 2025

• Fix VideoDecoder note in the docs

  #787 merged Jul 24, 2025

• Release 0.5

  #786 merged Jul 23, 2025

• Close avio object *before* encode destruction. And flush, too.

  #773 merged Jul 23, 2025

• Add stream_index seek mode, read frame index and update metadata

  #764 merged Jul 21, 2025

• Fix typing error

  #783 merged Jul 21, 2025

• Add tutorial for parallel decoding

  #778 merged Jul 15, 2025

• Remove obsolete stream_index from decoder benchmark calls

  #780 merged Jul 14, 2025

• Remove timing code in CudaDeviceInterface.cpp

  #771 merged Jul 14, 2025

• Fix CUDA x FFmpeg4 tests jobs

  #774 merged Jul 14, 2025

• Refactor C++ library names

  #768 merged Jul 11, 2025

• Generalize pybind_ops module name

  #759 merged Jul 10, 2025

• Use nullptr instead of NULL

  #763 merged Jul 10, 2025

9 Pull requests opened by 3 people

• Support 10 bits videos on CUDA

  #790 opened Jul 27, 2025

• [WIP] Fix full-range CUDA decoding

  #791 opened Jul 27, 2025

• [WIP] native nvcuvid backend

  #798 opened Jul 31, 2025

• [wip] Update VideoDecoder init

  #799 opened Aug 1, 2025

• [wip] Add npp_stream_ctx cache

  #800 opened Aug 1, 2025

• Test Windows CPU wheel

  #806 opened Aug 5, 2025

• Use MSVC toolchain when building FFmpeg on Windows

  #807 opened Aug 5, 2025

• Update version and README version table

  #809 opened Aug 7, 2025

• Tests: fix tests if running with self-built ffmpeg

  #812 opened Aug 8, 2025

11 Issues closed by 5 people

• Does `torchcodec` handle duplicate frame indices internally?

  #805 closed Aug 5, 2025

• Package Imports on src directory

  #804 closed Aug 5, 2025

• RuntimeError: Expected format to be AV_PIX_FMT_CUDA, got yuv420p

  #803 closed Aug 5, 2025

• Use pyav instead of ffmpeg, for better distribution

  #802 closed Aug 3, 2025

• Could not run 'torchcodec_ns::create_from_tensor' with arguments from the 'CPU' backend.

  #795 closed Aug 1, 2025

• Getting version torchcodec-0.0.0.dev0 installed on Windows

  #796 closed Jul 31, 2025

• Pixel histogram from GPU decoding shows unnatural trend and abnormal spikes

  #736 closed Jul 28, 2025

• noise when decoding hevc yuv420p10le videos with `device="cuda"`

  #598 closed Jul 28, 2025

• get_frames_by_pts() RuntimeError: Expected at most 2 arguments but received 3

  #779 closed Jul 14, 2025

• FFmpeg 4 fails to install in CI

  #734 closed Jul 14, 2025

• RuntimeError: Could not load libtorchcodec

  #762 closed Jul 10, 2025

9 Issues opened by 8 people

• Make samplers work for AudioDecoders

  #811 opened Aug 7, 2025

• Support for Passing Codec-Specific Options

  #810 opened Aug 7, 2025

• Intermittent green frames

  #801 opened Aug 1, 2025

• Proper 10bit videos support on GPU

  #776 opened Jul 13, 2025

• Use `ptsToSeconds()` consistently in C++ code

  #770 opened Jul 11, 2025

• Attempt to move Mac CI jobs to macos-m2-15

  #769 opened Jul 11, 2025

• Migrate all uses of `at::Tensor` to `torch::Tensor`.

  #767 opened Jul 10, 2025

• Multiprocessing leads to decoding error

  #766 opened Jul 10, 2025

• Unexpected benchmark results: torchcodec with GPU slower than CPU in some scenarios

  #765 opened Jul 10, 2025

5 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Support encoding to file-like object

  #754 commented on Aug 8, 2025 • 10 new comments

• Aarch64 support

  #569 commented on Jul 10, 2025 • 0 new comments

• Windows Support

  #640 commented on Aug 3, 2025 • 0 new comments

• torchcodec not compatible with brew-installed ffmpeg

  #570 commented on Aug 7, 2025 • 0 new comments

• Enable Intel GPU support in torchcodec on Linux (xpu device)

  #558 commented on Aug 8, 2025 • 0 new comments