What's Changed
New features
Turn on/off classification augmentations (#4039)
Turn on/off detection and instance segmentation augmentations (#4066)
Add GPU memory monitor hook (#4118)
Add YOLOv9 model for Object Detection (#3917, #4026)
Add OV inference for keypoint detection (#3970)
Add tiling for semantic segmentation (#3954)
Add D-Fine Detection Algorithm (#4142)
Enhancements
Update visual prompting pipeline for multi-label zero-shot learning support (#3993)
Update to work torch compile in detection (#4003)
Refactor MaskDINO (#4006)
Fix MaskRCNN/RTMDet-Inst/MaskRCNNTV Explain Mode (#4053)
Reduce tiling unittests resource consumption (#4073)
Upgrade OpenVINO to 2024.5 and NNCF to 2.14.0 (#4123)
Upgrade OV, MAPI, and NNCF dependencies (#3967)
Instance Segmentation Model refactoring (#3865)
Bump torch and lightning to 2.4.0 versions (#3843)
Add mAP metric to evaluate multilabel classification (#3985)
Bump OV to 2024.6, update empty label handling (#4162)
Bump MAPI to 0.2.5.1 (#4211)
Revert FMetric computation (PR: 4130) (#4210)
Bug fixes
Fix MaskDINO recipe (#4080)
Fix a wrong HPO log (#3972)
Update model name in rotated detection recipes (#4028)
Fix SupCon flag (#4076)
Add h-cls label info normalization (#4173)
Fix arrow support for semantic segmentation task (#4172)
Fix backward compatibility issues in model checkpoint loading (#4199)
Fix label info on loading checkpoint (#4200)
Revert h-cls head to linear one (#4221)
Full Changelog: 2.2.2...2.3.0