v0.5.0 (2/3/2023)

Highlights

1. Support RTMDet-R rotated object detection
2. Support for using mask annotation to improve YOLOv8 object detection performance
3. Support MMRazor searchable NAS sub-network as the backbone of YOLO series algorithm
4. Support calling MMRazor to distill the knowledge of RTMDet
5. MMYOLO document structure optimization, comprehensive content upgrade
6. Improve YOLOX mAP and training speed based on RTMDet training hyperparameters
7. Support calculation of model parameters and FLOPs, provide GPU latency data on T4 devices, and update Model Zoo
8. Support test-time augmentation (TTA)
9. Support RTMDet, YOLOv8 and YOLOv7 assigner visualization

New Features

1. Support inference for RTMDet instance segmentation tasks (#583)
2. Beautify the configuration file in MMYOLO and add more comments (#501, #506, #516, #529, #531, #539)
3. Refactor and optimize documentation (#568, #573, #579, #584, #587, #589, #596, #599, #600)
4. Support fast version of YOLOX (#518)
5. Support DeepStream in EasyDeploy and add documentation (#485, #545, #571)
6. Add confusion matrix drawing script (#572)
7. Add single channel application case (#460)
8. Support auto registration (#597)
9. Support Box CAM of YOLOv7, YOLOv8 and PPYOLOE (#601)
10. Add automated generation of MM series repo registration information and tools scripts (#559)
11. Added YOLOv7 model structure diagram (#504)
12. Add how to specify specific GPU training and inference files (#503)
13. Add check if metainfo is all lowercase when training or testing (#535)
14. Add links to Twitter, Discord, Medium, YouTube, etc. (#555)

Bug Fixes

1. Fix isort version issue (#492, #497)
2. Fix type error of assigner visualization (#509)
3. Fix YOLOv8 documentation link error (#517)
4. Fix RTMDet Decoder error in EasyDeploy (#519)
5. Fix some document linking errors (#537)
6. Fix RTMDet-Tiny weight path error (#580)

Improvements

1. Update contributing.md
2. Optimize DetDataPreprocessor branch to support multitasking (#511)
3. Optimize gt_instances_preprocess so it can be used for other YOLO algorithms (#532)
4. Add yolov7-e6e weight conversion script (#570)
5. Reference YOLOv8 inference code modification PPYOLOE

Contributors

A total of 23 developers contributed to this release.

Thank @triple-Mu, @isLinXu, @Audrey528, @TianWen580, @yechenzhi, @RangeKing, @lyviva, @Nioolek, @PeterH0323, @tianleiSHI, @aptsunny, @satuoqaq, @vansin, @xin-li-67, @VoyagerXvoyagerx, @landhill, @kitecats, @tang576225574, @HIT-cwh, @AI-Tianlong, @RangiLyu, @hhaAndroid, @liuyanyi