v0.3.0 (8/1/2023)

Highlights

1. Implement fast version of RTMDet. RTMDet-s 8xA100 training takes only 14 hours. The training speed is 2.6 times faster than the previous version.
2. Support PPYOLOE training
3. Support iscrowd attribute training in YOLOv5
4. Support YOLOv5 assigner result visualization

New Features

1. Add crowdhuman dataset (#368)
2. Easydeploy support TensorRT inference (#377)
3. Add YOLOX structure description (#402)
4. Add a feature for the video demo (#392)
5. Support YOLOv7 easy deploy (#427)
6. Add resume from specific checkpoint in CLI (#393)
7. Set metainfo fields to lower case (#362, #412)
8. Add module combination doc (#349, #352, #345)
9. Add docs about how to freeze the weight of backbone or neck (#418)
10. Add don't used pre-training weights doc in how_to.md (#404)
11. Add docs about how to set the random seed (#386)
12. Translate rtmdet_description.md document to English (#353)
13. Add doc of yolov6_description.md (#382, #372)

Bug Fixes

1. Fix bugs in the output annotation file when --class-id-txt is set (#430)
2. Fix batch inference bug in YOLOv5 head (#413)
3. Fix typehint in some heads (#415, #416, #443)
4. Fix RuntimeError of torch.cat() expected a non-empty list of Tensors (#376)
5. Fix the device inconsistency error in YOLOv7 training (#397)
6. Fix the scale_factor and pad_param value in LetterResize (#387)
7. Fix docstring graph rendering error of readthedocs (#400)
8. Fix AssertionError when YOLOv6 from training to val (#378)
9. Fix CI error due to np.int and legacy builder.py (#389)
10. Fix MMDeploy rewriter (#366)
11. Fix MMYOLO unittest scope bug (#351)
12. Fix pad_param error (#354)
13. Fix twice head inference bug (#342)
14. Fix customize dataset training (#428)

Improvements

1. Update useful_tools.md (#384)
2. update the English version of custom_dataset.md (#381)
3. Remove context argument from the rewriter function (#395)
4. deprecating np.bool type alias (#396)
5. Add new video link for custom dataset (#365)
6. Export onnx for model only (#361)
7. Add MMYOLO regression test yml (#359)
8. Update video tutorials in article.md (#350)
9. Add deploy demo (#343)
10. Optimize the vis results of large images in debug mode (#346)
11. Improve args for browse_dataset and support RepeatDataset (#340, #338)

Contributors

A total of 28 developers contributed to this release.

Thank @RangeKing, @PeterH0323, @Nioolek, @triple-Mu, @matrixgame2018, @xin-li-67, @tang576225574, @kitecats, @Seperendity, @diplomatist, @vaew, @wzr-skn, @VoyagerXvoyagerx, @MambaWong, @tianleiSHI, @caj-github, @zhubochao, @lvhan028, @dsghaonan, @lyviva, @yuewangg, @wang-tf, @satuoqaq, @grimoire, @RunningLeon, @hanrui1sensetime, @RangiLyu, @hhaAndroid