XiaoHongbo-Hope opened a new pull request, #9831: URL: https://github.com/apache/paimon/pull/9831
### Purpose Exact random video reads currently scan the packet stream when a decoder first opens. Persist the exact PTS, duration, and key-frame mapping once during ingestion so repeated training reads can seek without that scan. This complements #9799, which optimizes decoding after the decoder is open. TorchCodec provides the same concept through `custom_frame_mappings`: https://meta-pytorch.org/torchcodec/stable/generated_examples/decoding/custom_frame_mappings.html ### Changes - Add optional per-video frame mappings to `.video` v2 and `VideoFrameDescriptor` v2. - Let TorchCodec and PyAV reuse the mapping for exact frame access. - Preserve mappings during Java/Python reads and compaction. - Keep `.video` v1 readable; missing mappings use the existing scan fallback. ### Tests - 71 Python video/LeRobot tests passed, 28 skipped. - Java `VideoFrameDescriptorTest` and `VideoFileFormatTest` passed. - Python 3.6 syntax compilation, flake8, Spotless, and Checkstyle passed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
