XiaoHongbo-Hope opened a new pull request, #10405: URL: https://github.com/apache/paimon/pull/10405
`PaimonLeRobotDataset` currently copies decoded video windows once per sample and again during DataLoader collation. Add `to_dataloader()` to assemble video windows directly in worker-shared batches while preserving the existing indexing API and the per-sample `image_transforms` path. CPU TorchCodec benchmark (4 cameras × 8 frames, batch 32, 8 workers, 3 paired seeds): **164.94 → 220.47 samples/s (+33.67%)**. Median batch p95 falls from 962.1 ms to 483.8 ms; sample-order and image hashes match in every pair. Tests: `multimodal_lerobot_test.py` — 66 passed, 29 skipped; Python 3.6 compile, flake8, and `git diff --check` pass. -- 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]
