v1.0.0rc5(30/12/2022)

Highlights

- Support EVA, RevViT, EfficientnetV2, CLIP, TinyViT and MixMIM backbones.
- Reproduce the training accuracy of ConvNeXt and RepVGG.
- Support multi-task training and testing.
- Support Test-time Augmentation.

New Features

- [Feature] Add EfficientnetV2 Backbone. ([open-mmlab#1253](open-mmlab#1253))
- [Feature] Support TTA and add `--tta` in `tools/test.py`. ([open-mmlab#1161](open-mmlab#1161))
- [Feature] Support Multi-task. ([open-mmlab#1229](open-mmlab#1229))
- [Feature] Add clip backbone. ([open-mmlab#1258](open-mmlab#1258))
- [Feature] Add mixmim backbone with checkpoints. ([open-mmlab#1224](open-mmlab#1224))
- [Feature] Add TinyViT for dev-1.x. ([open-mmlab#1042](open-mmlab#1042))
- [Feature] Add some scripts for development. ([open-mmlab#1257](open-mmlab#1257))
- [Feature] Support EVA. ([open-mmlab#1239](open-mmlab#1239))
- [Feature] Implementation of RevViT. ([open-mmlab#1127](open-mmlab#1127))

Improvements

- [Reproduce] Reproduce RepVGG  Training Accuracy. ([open-mmlab#1264](open-mmlab#1264))
- [Enhance] Support ConvNeXt More Weights. ([open-mmlab#1240](open-mmlab#1240))
- [Reproduce] Update ConvNeXt config files. ([open-mmlab#1256](open-mmlab#1256))
- [CI] Update CI to test PyTorch 1.13.0. ([open-mmlab#1260](open-mmlab#1260))
- [Project] Add ACCV workshop 1st Solution. ([open-mmlab#1245](open-mmlab#1245))
- [Project] Add Example project. ([open-mmlab#1254](open-mmlab#1254))

Bug Fixes

- [Fix] Fix imports in transforms. ([open-mmlab#1255](open-mmlab#1255))
- [Fix] Fix CAM visualization. ([open-mmlab#1248](open-mmlab#1248))
- [Fix] Fix the requirements and lazy register mmcls models. ([open-mmlab#1275](open-mmlab#1275))

v1.0.0rc4(06/12/2022)

Highlights

- Upgrade API to get pre-defined models of MMClassification. See [open-mmlab#1236](open-mmlab#1236) for more details.
- Refactor BEiT backbone and support v1/v2 inference. See [open-mmlab#1144](open-mmlab#1144).

New Features

- Support getting model from the name defined in the model-index file. ([open-mmlab#1236](open-mmlab#1236))

Improvements

- Support evaluate on both EMA and non-EMA models. ([open-mmlab#1204](open-mmlab#1204))
- Refactor BEiT backbone and support v1/v2 inference. ([open-mmlab#1144](open-mmlab#1144))

Bug Fixes

- Fix `reparameterize_model.py` doesn't save meta info. ([open-mmlab#1221](open-mmlab#1221))
- Fix dict update in BEiT. ([open-mmlab#1234](open-mmlab#1234))

Docs Update

- Update install tutorial. ([open-mmlab#1223](open-mmlab#1223))
- Update MobileNetv2 & MobileNetv3 readme. ([open-mmlab#1222](open-mmlab#1222))
- Add version selection in the banner. ([open-mmlab#1217](open-mmlab#1217))

v0.25.0(06/12/2022)

Highlights

- Support MLU backend.

New Features

- Support MLU backend. ([open-mmlab#1159](open-mmlab#1159))
- Support Activation Checkpointing for ConvNeXt. ([open-mmlab#1152](open-mmlab#1152))

Improvements

- Add `dist_train_arm.sh` for ARM device and update NPU results. ([open-mmlab#1218](open-mmlab#1218))

Bug Fixes

- Fix a bug caused `MMClsWandbHook` stuck. ([open-mmlab#1242](open-mmlab#1242))
- Fix the redundant `device_ids` in `tools/test.py`. ([open-mmlab#1215](open-mmlab#1215))

Docs Update

- Add version banner and version warning in master docs. ([open-mmlab#1216](open-mmlab#1216))
- Update NPU support doc. ([open-mmlab#1198](open-mmlab#1198))
- Fixed typo in `pytorch2torchscript.md`. ([open-mmlab#1173](open-mmlab#1173))
- Fix typo in `miscellaneous.md`. ([open-mmlab#1137](open-mmlab#1137))
- further detail for the doc for `ClassBalancedDataset`. ([open-mmlab#901](open-mmlab#901))

v1.0.0rc3(21/11/2022)

Highlights

- Add **Switch Recipe** Hook, Now we can modify training pipeline, mixup and loss settings during training, see [open-mmlab#1101](open-mmlab#1101).
- Add **TIMM and HuggingFace** wrappers. Now you can train/use models in TIMM/HuggingFace directly, see [open-mmlab#1102](open-mmlab#1102).
- Support **retrieval tasks**, see [open-mmlab#1055](open-mmlab#1055).
- Reproduce **mobileone** training accuracy. See [open-mmlab#1191](open-mmlab#1191)

New Features

- Add checkpoints from EfficientNets NoisyStudent & L2. ([open-mmlab#1122](open-mmlab#1122))
- Migrate CSRA head to 1.x. ([open-mmlab#1177](open-mmlab#1177))
- Support RepLKnet backbone. ([open-mmlab#1129](open-mmlab#1129))
- Add Switch Recipe Hook. ([open-mmlab#1101](open-mmlab#1101))
- Add adan optimizer. ([open-mmlab#1180](open-mmlab#1180))
- Support DaViT. ([open-mmlab#1105](open-mmlab#1105))
- Support Activation Checkpointing for ConvNeXt. ([open-mmlab#1153](open-mmlab#1153))
- Add TIMM and HuggingFace wrappers to build classifiers from them directly. ([open-mmlab#1102](open-mmlab#1102))
- Add reduction for neck ([open-mmlab#978](open-mmlab#978))
- Support HorNet Backbone for dev1.x. ([open-mmlab#1094](open-mmlab#1094))
- Add arcface head. ([open-mmlab#926](open-mmlab#926))
- Add Base Retriever and Image2Image Retriever for retrieval tasks. ([open-mmlab#1055](open-mmlab#1055))
- Support MobileViT backbone. ([open-mmlab#1068](open-mmlab#1068))

Improvements

- [Enhance] Enhance ArcFaceClsHead. ([open-mmlab#1181](open-mmlab#1181))
- [Refactor] Refactor to use new fileio API in MMEngine. ([open-mmlab#1176](open-mmlab#1176))
- [Enhance] Reproduce mobileone training accuracy. ([open-mmlab#1191](open-mmlab#1191))
- [Enhance] add deleting params info in swinv2. ([open-mmlab#1142](open-mmlab#1142))
- [Enhance] Add more mobilenetv3 pretrains. ([open-mmlab#1154](open-mmlab#1154))
- [Enhancement] RepVGG for YOLOX-PAI for dev-1.x. ([open-mmlab#1126](open-mmlab#1126))
- [Improve] Speed up data preprocessor. ([open-mmlab#1064](open-mmlab#1064))

Bug Fixes

- Fix the torchserve. ([open-mmlab#1143](open-mmlab#1143))
- Fix configs due to api refactor of `num_classes`. ([open-mmlab#1184](open-mmlab#1184))
- Update mmcls2torchserve. ([open-mmlab#1189](open-mmlab#1189))
- Fix for `inference_model` cannot get classes information in checkpoint. ([open-mmlab#1093](open-mmlab#1093))

Docs Update

- Add not-found page extension. ([open-mmlab#1207](open-mmlab#1207))
- update visualization doc. ([open-mmlab#1160](open-mmlab#1160))
- Support sort and search the Model Summary table. ([open-mmlab#1100](open-mmlab#1100))
- Improve the ResNet model page. ([open-mmlab#1118](open-mmlab#1118))
- update the readme of convnext. ([open-mmlab#1156](open-mmlab#1156))
- Fix the installation docs link in README. ([open-mmlab#1164](open-mmlab#1164))
- Improve ViT and MobileViT model pages. ([open-mmlab#1155](open-mmlab#1155))
- Improve Swin Doc and Add Tabs enxtation. ([open-mmlab#1145](open-mmlab#1145))
- Add MMEval projects link in README. ([open-mmlab#1162](open-mmlab#1162))
- Add runtime configuration docs. ([open-mmlab#1128](open-mmlab#1128))
- Add custom evaluation docs ([open-mmlab#1130](open-mmlab#1130))
- Add custom pipeline docs. ([open-mmlab#1124](open-mmlab#1124))
- Add MMYOLO projects link in MMCLS1.x. ([open-mmlab#1117](open-mmlab#1117))

v0.24.1(31/10/2022)

New Features

- Support mmcls with NPU backend. ([open-mmlab#1072](open-mmlab#1072))

Bug Fixes

- Fix performance issue in convnext DDP train. ([open-mmlab#1098](open-mmlab#1098))

v1.0.0rc2(12/10/2022)

New Features

- Support DeiT3. ([open-mmlab#1065](open-mmlab#1065))

Improvements

- Update `analyze_results.py` for dev-1.x. ([open-mmlab#1071](open-mmlab#1071))
- Get scores from inference api. ([open-mmlab#1070](open-mmlab#1070))

Bug Fixes

- Update requirements. ([open-mmlab#1083](open-mmlab#1083))

Docs Update

- Add 1x docs schedule. ([open-mmlab#1015](open-mmlab#1015))

v1.0.0rc1(30/9/2022)

New Features

- Support MViT for MMCLS 1.x ([open-mmlab#1023](open-mmlab#1023))
- Add ViT huge architecture. ([open-mmlab#1049](open-mmlab#1049))
- Support EdgeNeXt for dev-1.x. ([open-mmlab#1037](open-mmlab#1037))
- Support Swin Transformer V2 for MMCLS 1.x. ([open-mmlab#1029](open-mmlab#1029))
- Add efficientformer Backbone for MMCls 1.x. ([open-mmlab#1031](open-mmlab#1031))
- Add MobileOne Backbone For MMCls 1.x.  ([open-mmlab#1030](open-mmlab#1030))
- Support BEiT Transformer layer. ([open-mmlab#919](open-mmlab#919))

Improvements

- \[Refactor\] Fix visualization tools. ([open-mmlab#1045](open-mmlab#1045))
- \[Improve\] Update benchmark scripts ([open-mmlab#1028](open-mmlab#1028))
- \[Improve\] Update tools to enable `pin_memory` and `persistent_workers` by default. ([open-mmlab#1024](open-mmlab#1024))
- \[CI\] Update circle-ci and github workflow. ([open-mmlab#1018](open-mmlab#1018))

Bug Fixes

- Fix verify dataset tool in 1.x. ([open-mmlab#1062](open-mmlab#1062))
- Fix `loss_weight` in `LabelSmoothLoss`. ([open-mmlab#1058](open-mmlab#1058))
- Fix the output position of Swin-Transformer. ([open-mmlab#947](open-mmlab#947))

Docs Update

- Auto generate model summary table.  ([open-mmlab#1010](open-mmlab#1010))
- Refactor new modules tutorial. ([open-mmlab#998](open-mmlab#998))

v0.24.0(30/9/2022)

Highlights

- Support HorNet, EfficientFormerm, SwinTransformer V2 and MViT backbones.
- Support Standford Cars dataset.

New Features

- Support HorNet Backbone. ([open-mmlab#1013](open-mmlab#1013))
- Support EfficientFormer. ([open-mmlab#954](open-mmlab#954))
- Support Stanford Cars dataset. ([open-mmlab#893](open-mmlab#893))
- Support CSRA head. ([open-mmlab#881](open-mmlab#881))
- Support Swin Transform V2. ([open-mmlab#799](open-mmlab#799))
- Support MViT and add checkpoints. ([open-mmlab#924](open-mmlab#924))

Improvements

- \[Improve\] replace loop of progressbar in api/test. ([open-mmlab#878](open-mmlab#878))
- \[Enhance\] RepVGG for YOLOX-PAI. ([open-mmlab#1025](open-mmlab#1025))
- \[Enhancement\] Update VAN. ([open-mmlab#1017](open-mmlab#1017))
- \[Refactor\] Re-write `get_sinusoid_encoding` from third-party implementation. ([open-mmlab#965](open-mmlab#965))
- \[Improve\] Upgrade onnxsim to v0.4.0. ([open-mmlab#915](open-mmlab#915))
- \[Improve\] Fixed typo in `RepVGG`. ([open-mmlab#985](open-mmlab#985))
- \[Improve\] Using `train_step` instead of `forward` in PreciseBNHook ([open-mmlab#964](open-mmlab#964))
- \[Improve\] Use `forward_dummy` to calculate FLOPS. ([open-mmlab#953](open-mmlab#953))

Bug Fixes

- Fix warning with `torch.meshgrid`. ([open-mmlab#860](open-mmlab#860))
- Add matplotlib minimum version requriments. ([open-mmlab#909](open-mmlab#909))
- val loader should not drop last by default. ([open-mmlab#857](open-mmlab#857))
- Fix config.device bug in toturial. ([open-mmlab#1059](open-mmlab#1059))
- Fix attenstion clamp max params ([open-mmlab#1034](open-mmlab#1034))
- Fix device mismatch in Swin-v2. ([open-mmlab#976](open-mmlab#976))
- Fix the output position of Swin-Transformer. ([open-mmlab#947](open-mmlab#947))

Docs Update

- Fix typo in config.md. ([open-mmlab#827](open-mmlab#827))
- Add version for torchvision to avoide error. ([open-mmlab#903](open-mmlab#903))
- Fixed typo for `--out-dir` option of analyze_results.py. ([open-mmlab#898](open-mmlab#898))
- Refine the docstring of RegNet ([open-mmlab#935](open-mmlab#935))

v1.0.0rc0(31/8/2022)

MMClassification 1.0.0rc0 is the first version of MMClassification 1.x, a part of the OpenMMLab 2.0 projects.

Built upon the new [training engine](https://github.com/open-mmlab/mmengine), MMClassification 1.x unifies the interfaces of dataset, models, evaluation, and visualization.

And there are some BC-breaking changes. Please check [the migration tutorial](https://mmclassification.readthedocs.io/en/1.x/migration.html) for more details.

v0.23.2(28/7/2022)

New Features

- Support MPS device. ([open-mmlab#894](open-mmlab#894))

Bug Fixes

- Fix a bug in Albu which caused crashing. ([open-mmlab#918](open-mmlab#918))