MMDetection V3.0.0rc6 Release

[ZwwWayne]

Highlights

• Support Boxinst, Objects365 Dataset, and Separated and Occluded COCO metric
• Support ConvNeXt-V2, DiffusionDet, and inference of EfficientDet and Detic in Projects
• Refactor DETR series and support Conditional-DETR, DAB-DETR, and DINO
• Support DetInferencer for inference, Test Time Augmentation, and automatically importing modules from registry
• Support RTMDet-Ins ONNXRuntime and TensorRT deployment
• Support calculating FLOPs of detectors

New Features

• Support Boxinst (CodeCamp #1492: Support Boxinst #9525)
• Support Objects365 Dataset ([Feature] Support Objects365 Dataset #9600)
• Support ConvNeXt-V2 in Projects ([Feature] Support ConvNeXt-V2. #9619)
• Support DiffusionDet in Projects ([Feature] Support DiffusionDet Inference in Projects #9639, [Feature] Fully support DiffusionDet in projects #9768)
• Support Detic inference in Projects ([Projects] Support Detic inference. #9645)
• Support EfficientDet inference in Projects ([Projects] Support Detic inference. #9645)
• Support Separated and Occluded COCO metric ([Feature] Migrate occluded metric to 3.x. #9710)
• Support auto import modules from registry ([Feature] Support auto import modules from registry. #9143)
• Refactor DETR series and support Conditional-DETR, DAB-DETR and DINO (Refactor DETRs and add C-DETR, DAB-DETR and DINO #9646)
• Support DetInferencer for inference ([Feature] Support DetInferencer #9561)
• Support Test Time Augmentation ([Feature] Support TTA #9452)
• Support calculating FLOPs of detectors ([Fix] Fix compute flops problems #9777)

Bug Fixes

• Fix deprecating old type alias due to new version of numpy (Update loading.py #9625, [Fix] Deprecating old type alias due to new version of numpy #9537)
• Fix VOC metrics ([Fix] Fix VOC metrics #9784)
• Fix the wrong link of RTMDet-x log ([Fix] the wrong link of RTMDet-x log #9549)
• Fix RTMDet link in README (fix rtmdet link #9575)
• Fix MMDet get flops error ([Fix] Fix mmdet flops error #9589)
• Fix use_depthwise in RTMDet ([Fix] Fix use_depthwise in RTMDet. #9624)
• Fix albumentations augmentation post process with masks ([Fix] albu augmentation post process with masks #9551)
• Fix DETR series Unit Test (fix ci #9647)
• Fix LoadPanopticAnnotations bug (【Fix】fix ci test bug #9703)
• Fix isort CI (Fix isort ci #9680)
• Fix amp pooling overflow (fix amp pooling overflow #9670)
• Fix docstring about noise in DINO ([Fix] Fix docstring about noise in DINO #9747)
• Fix potential bug in MultiImageMixDataset ([Fix] Fix potential bug in MultiImageMixDataset #9764)

Improvements

• Replace NumPy transpose with PyTorch permute to speed-up (Replace numpy transpose with torch permute to speed-up #9762)
• Deprecate sklearn (RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same #9725)
• Add RTMDet-Ins deployment guide ([Doc] Add RTMDet-Ins deployment guide. #9823)
• Update RTMDet config and README ([Enhance] Update rtmdet config and readme. #9603)
• Replace the models used in the tutorial document with RTMDet ([Refactor] Refactor LVIS metric with MMEval. #9843)
• Adjust the minimum supported python version to 3.7 ([Fix] Adjust the minimum supported python version to 3.7 #9602)
• Support modifying palette through configuration ([Enhance] Support modifying palette through configuration #9445)
• Update README document in Project ([Enhance] Update README document in Project #9599)
• Replace github with gitee in .pre-commit-config-zh-cn.yaml file ([pre-commit] Replace github with gitee #9586)
• Use official isort in .pre-commit-config.yaml file (dev-3.x: Use official isort #9701)
• Change MMCV minimum version to 2.0.0rc4 for dev-3.x ([Fix]: Change MMCV minimum version to 2.0.0rc4 for dev-3.x branch #9695)
• Add Chinese version of single_stage_as_rpn.md and test_results_submission.md (CodeCamp #34 [Docs] Add Chinese version of single_stage_as_rpn.md and test_results_submission.md #9434)
• Add OpenDataLab download link ([Doc] add opendatalab download link #9605, [Doc] update opendatalab download link (add postfix) #9738)
• Add type hints of several layers (CodeCamp #33 [Fix]:add type hints for res_layer, se_layer,normed_predictor,positional_encoding #9346)
• Add typehint for DarknetBottleneck ([Fix] Add typehint for DarknetBottleneck #9591)
• Add dockerfile that is easier to use in China (add dockerfile #9659)
• Add twitter, discord, medium, and youtube link (docs: Add twitter discord medium youtube link #9775)
• Prepare for merging refactor-detr ([Fix] Prepare for merging refactor-detr #9656)
• Add metafile to ConditionalDETR, DABDETR and DINO (Add metafile to ConditionalDETR DABDETR and DINO #9715)
• Support to modify non_blocking parameters (Support to modify non_blocking parameters #9723)
• Comment repeater visualizer register (comment repeater visualizer register #9740)
• Update user guide: finetune.md and inference.md ([Doc]: update user guide: "finetune.md, inference.md" #9578)

New Contributors

• @NoFish-528 made their first contribution in CodeCamp #33 [Fix]:add type hints for res_layer, se_layer,normed_predictor,positional_encoding #9346
• @137208 made their first contribution in CodeCamp #34 [Docs] Add Chinese version of single_stage_as_rpn.md and test_results_submission.md #9434
• @lyviva made their first contribution in Update loading.py #9625
• @zwhus made their first contribution in [Fix] Fix mmdet flops error #9589
• @zylo117 made their first contribution in fix amp pooling overflow #9670
• @chg0901 made their first contribution in comment repeater visualizer register #9740
• @DanShouzhu made their first contribution in [Doc]: update user guide: "finetune.md, inference.md" #9578

Contributors

A total of 27 developers contributed to this release.

Thanks @JosonChan1998, @RangeKing, @NoFish-528, @likyoo, @Xiangxu-0103, @137208, @PeterH0323, @tianleiSHI, @wufan-tb, @lyviva, @zwhus, @jshilong, @Li-Qingyun, @sanbuphy, @zylo117, @triple-Mu, @KeiChiTse, @LYMDLUT, @nijkah, @chg0901, @DanShouzhu, @zytx121, @vansin, @BIGWangYuDong, @hhaAndroid, @RangiLyu, @ZwwWayne

Full Changelog: v3.0.0rc5...v3.0.0rc6

---

This discussion was created from the release MMDetection V3.0.0rc6 Release. Feel free to leave messages to give suggestions or feedback on this version!