Name | Modified | Size | Downloads / Week |
---|---|---|---|
Parent folder | |||
MMAction2 V1.0.0 Release.tar.gz | 2023-04-07 | 10.2 MB | |
MMAction2 V1.0.0 Release.zip | 2023-04-07 | 10.8 MB | |
README.md | 2023-04-07 | 2.9 kB | |
Totals: 3 Items | 21.1 MB | 0 |
Highlights
- Support RGB-PoseC3D(CVPR'2022).
- Support training UniFormer V2(Arxiv'2022).
- Support MSG3D(CVPR'2020) and CTRGCN(CVPR'2021) in projects.
- Refactor and provide more user-friendly documentation.
New Features
- Support RGB-PoseC3D (2182)
- Support training UniFormer V2 (2221)
- Support MSG3D and CTRGCN in projects. (2269, 2291)
Improvements
- Use MMEngine to calculate FLOPs (2300)
- Speed up LFB training (2294)
- Support multiprocessing on AVA evaluation (2146)
- Add a demo for exporting spatial-temporal detection model to ONNX (2225)
- Update spatial-temporal detection related folders (2262)
Bug Fixes
- Fix flip config of TSM for sth v1/v2 dataset (#2247)
- Fix circle ci (2336, 2334)
- Fix accepting an unexpected argument local-rank in PyTorch 2.0 (2320)
- Fix TSM config link (2315)
- Fix numpy version requirement in CI (2284)
- Fix NTU pose extraction script (2246)
- Fix TSM-MobileNet V2 (2332)
- Fix command bugs in localization tasks' README (2244)
- Fix duplicate name in DecordInit and SampleAVAFrame (2251)
- Fix channel order when showing video (2308)
- Specify map_location to cpu when using _load_checkpoint (2252)
Documentation