Repository navigation
[python] Speed up TorchCodec random video frame reads with persisted indexes - #10821
Open
XiaoHongbo-Hope wants to merge 7 commits into
Open
XiaoHongbo-Hope wants to merge 7 commits into
XiaoHongbo-Hope wants to merge 7 commits into
Conversation
XiaoHongbo-Hope
changed the title
XiaoHongbo-Hope
changed the title
XiaoHongbo-Hope
marked this pull request as ready for review
October 31, 2026 15:51
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
Explicit TorchCodec reads currently scan videos during initialization, even when a persisted keyframe index is available.
Use the MP4 sample table for timestamp seeks, prefetch metadata and target GOP ranges, and verify returned PTS. Unsupported layouts or failed verification fall back to exact mode. Applies to
video_backend="torchcodec"; the default backend is unchanged.Replica decoders share read-only timelines. Uniform timing uses arithmetic sequences; irregular timing retains one shared pair of arrays. Metadata is released with the last decoder and is not transferred to workers. Keyframe lookup no longer copies an ordinal list.
Tests
57 tests passed with real PyAV/TorchCodec, covering B-frames, VFR, track selection, fallback, million-sample memory bounds, replica sharing and cleanup. Worker cache reset, Python 3.6 syntax, flake8 and diff checks passed.