Highlights

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

New Features

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

Improvements

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

Bug Fixes

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

Docs Update

• Add version for torchvision to avoid error. (#903)
• Fix typo for --out-dir option of analyze_results.py. (#898)
• Refine the docstring of RegNet. (#935)

Contributors

A total of 19 developers contributed to this release.

@a-mos @Ezra-Yu @Fei-Wang @nijkah @PeterH0323 @yingfhu @techmonsterwang @JiayuXu0 @jlim262 @hukkai @mzr1996 @liu-mengyang @twmht @pallgeuer @timothylimyl @daquexian @okotaku @tpoisonooo @zzc98

Highlights

• Support MViT, EdgeNeXt, Swin-Transformer V2, EfficientFormer and MobileOne.
• Support BEiT type transformer layer.

New Features

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

Improvements

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

Bug Fixes

• Fix verify dataset tool in 1.x. (#1062)
• Fix loss_weight in LabelSmoothLoss. (#1058)
• Fix the output position of Swin-Transformer. (#947)

Docs Update

• Fix typo in migration document. (#1063)
• Auto generate model summary table. (#1010)
• Refactor new modules tutorial. (#998)

Contributors

A total of 8 developers contributed to this release.

@Ezra-Yu @yingfhu @mzr1996 @tonysy @fangyixiao18 @YuanLiuuuuuu @HIT-cwh @techmonsterwang

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, MMClassification 1.x unifies the interfaces of dataset, models, evaluation, and visualization.

And there are some BC-breaking changes. Please check the migration tutorial for more details.

New Features

• Support MPS device. (#894)

Improvements

• Add test mim CI. (#879)

Bug Fixes

• [Fix] Fix Albu crash bug. (#918)
• [Fix] Add mim to extras_require in setup.py. (#872)

Contributors

A total of 2 developers contributed to this release.

@mzr1996 @PeterH0323

Highlights

• New WandbHook to store your training log and visualize validation results!

New Features

• [Feature] Dedicated MMClsWandbHook for MMClassification (Weights and Biases Integration) (#764)

Improvements

• [Refactor] Use mdformat instead of markdownlint to format markdown. (#844)

Bug Fixes

• [Fix] Fix wrong --local_rank.

Docs Update

• [Docs] Update install tutorials. (#854)
• [Docs] Fix wrong link in README. (#835)

Contributors

A total of 3 developers contributed to this release.

@ayulockin @mzr1996 @timothylimyl

New Features

• Support DenseNet. (#750)
• Support VAN. (#739)

Improvements

• Support training on IPU and add fine-tuning configs of ViT. (#723)

Docs Update

• New style API reference, and easier to use! Welcome view it. (#774)

Contributors

A total of 4 developers contributed to this release.

@mzr1996 @okotaku @yingfhu @HuDi2018

New Features

• Support resize relative position embedding in SwinTransformer. (#749)
• Add PoolFormer backbone and checkpoints. (#746)

Improvements

• Improve CPE performance by reduce memory copy. (#762)
• Add extra dataloader settings in configs. (#752)

Contributors

A total of 4 developers contributed to this release.

@mzr1996 @yuweihao @XiaobingSuper @YuanLiuuuuuu

Considering more and more codebase depends on new features of MMClassification, we will release a minor version at the middle of every month. 😉

Highlights

• Support a series of CSP Network, such as CSP-ResNet, CSP-ResNeXt and CSP-DarkNet.
• A new CustomDataset class to help you build dataset of yourself!
• Support ConvMixer, RepMLP and new dataset - CUB dataset.

New Features

• Add CSPNet and backbone and checkpoints (#735)
• Add CustomDataset. (#738)
• Add diff seeds to diff ranks. (#744)
• Support ConvMixer. (#716)
• Our dist_train & dist_test tools support distributed training on multiple machines. (#734)
• Add RepMLP backbone and checkpoints. (#709)
• Support CUB dataset. (#703)
• Support ResizeMix. (#676)

Improvements

• Use --a-b instead of --a_b in arguments. (#754)
• Add get_cat_ids and get_gt_labels to KFoldDataset. (#721)
• Set torch seed in worker_init_fn. (#733)

Bug Fixes

• Fix the discontiguous output feature map of ConvNeXt. (#743)

Docs Update

• Add brief installation steps in README for copy&paste. (#755)
• fix logo url link from mmocr to mmcls. (#732)

Contributors

A total of 6 developers contributed to this release.

@Ezra-Yu @yingfhu @Hydrion-Qlz @mzr1996 @huyu398 @okotaku

Highlights

• Support ResNetV1c and Wide-ResNet, and provide pre-trained models.
• Support dynamic input shape for ViT-based algorithms. Now our ViT, DeiT, Swin-Transformer and T2T-ViT supports forwarding with any input shape.
• Reproduce training results of DeiT. And our DeiT-T and DeiT-S have higher accuracy comparing with the official weights.

New Features

• Add ResNetV1c. (#692)
• Support Wide-ResNet. (#715)
• Support gem pooling (#677)

Improvements

• Reproduce training results of DeiT. (#711)
• Add ConvNeXt pretrain models on ImageNet-1k. (#707)
• Support dynamic input shape for ViT-based algorithms. (#706)
• Add evaluate function for ConcatDataset. (#650)
• Enhance vis-pipeline tool. (#604)
• Return code 1 if scripts runs failed. (#694)
• Use PyTorch official one_hot to implement convert_to_one_hot. (#696)
• Add a new pre-commit-hook to automatically add a copyright. (#710)
• Add deprecation message for deploy tools. (#697)
• Upgrade isort pre-commit hooks. (#687)
• Use --gpu-id instead of --gpu-ids in non-distributed multi-gpu training/testing. (#688)
• Remove deprecation. (#633)

Bug Fixes

• Fix Conformer forward with irregular input size. (#686)
• Add dist.barrier to fix a bug in directory checking. (#666)

Contributors

A total of 8 developers contributed to this release.

@Ezra-Yu @HumberMe @mzr1996 @twmht @RunningLeon @yasu0001 @okotaku @yingfhu

Bug Fixes

• Fix the MMCV dependency version.