We are excited to announce the release of MMSelfSup v1.0.0rc0. MMSelfSup v1.0.0rc0 is the first version of MMSelfSup 1.x, a part of the OpenMMLab 2.x projects. The master branch is still 0.x version and we will checkout a new 1.x branch to release 1.x version. The two versions will be maintained simultaneously in the future.

We briefly list the major breaking changes here. Please refer to the migration guide for details and migration instructions.

Highlight

• Based on MMEngine and MMCV.
• Released with refactor.
  • Datasets
  • Models
  • Config
  • ...
• Refine all documents.

New Features

• Add SelfSupDataSample to unify the components' interface.
• Add SelfSupVisualizer for visualization.
• Add SelfSupDataPreprocessor for data preprocess in model.

Improvements

• Most algorithms now support non-distributed training.
• Change the interface of different data augmentation transforms to dict.
• Run classification downstream task with MMClassification.

Docs

• Refine all documents and reorganize the directory.
• Add concepts for different components.

New Features

• Support MAE Reconstructed Image Visualization (#376)

Bug Fixes

• Fix args/cfg bug in extract.py, use cfg.work_dir to save files (#357)
• Fix SimMIM mask generator config bug (#360)

Improvements

• Update mdformat settings (#323)
• Add circleci (#374)

Docs

• Fix the link of switch language (#327)
• Update lr_updater.py links in tutorials/4_schedule.md (#354)

Contributors

A total of 4 developers contributed to this release. Thanks @fangyixiao18 , @YuanLiuuuuuu , @soonera , @doem97

Highlight

• Update BYOL model and results (#319)
• Refine some documentation

New Features

• Update BYOL models and results (#319)

Bug Fixes

• Set qkv bias to False for cae and True for mae (#303)
• Fix spelling errors in MAE config (#307)

Improvements

• Change the file name of cosine annealing hook (#304)
• Replace markdownlint with mdformat (#311)

Docs

• Fix typo in tutotial (#308)
• Configure Myst-parser to parse anchor tag (#309)
• Update readthedocs algorithm README (#310)
• Rewrite install.md (#317)
• refine README.md file (#318)

Contributors

A total of 5 developers contributed to this release. Thanks @Jiahao000 , @fangyixiao18 , @YuanLiuuuuuu , @luuuyi , @soonera

Highlight

• Support CAE (#284)
• Support Barlow Twins (#207)

New Features

• Support CAE (#284)
• Support Barlow twins (#207)
• Add SimMIM 192 pretrain and 224 fine-tuning results (#280)
• Add MAE pretrain with fp16 (#271)

Bug Fixes

• Fix args error ([#290]([Fix] fix args error by fangyixiao18 · Pull Request #290 · open-mmlab/mmselfsup))
• Change imgs_per_gpu to samples_per_gpu in MAE config (#278)
• Avoid GPU memory leak with prefetch dataloader ([#277]([Feature] Avoid GPU memory leak with prefetch dataloader by mitming · Pull Request #277 · open-mmlab))
• Fix key error bug when registering custom hooks ([#273]([Fix] fix key error bug when registering custom hooks by fangyixiao18 · Pull Request #273 · open-mml))

Improvements

• Update SimCLR models and results (#295)
• Reduce memory usage while running unit test (#291)
• Remove PyTorch1.5 test (#288)
• Rename linear probing config file names ([#281]([Benchmark] rename linear probing config file names by fangyixiao18 · Pull Request #281 · open-mmlab))
• add unit test for apis ([#276]([UT] add unit test for apis by fangyixiao18 · Pull Request #276 · open-mmlab/mmselfsup))

Docs

• Fix SimMIM config link, and add SimMIM to model_zoo (#272)

Contributors

A total of 5 developers contributed to this release. Thanks @Jiahao000 , @fangyixiao18 , @YuanLiuuuuuu , @mitming , @xcnick

Highlight

• Support SimMIM (#239)
• Add KNN benchmark, support KNN test with checkpoint and extracted backbone weights (#243)
• Support ImageNet-21k dataset (#225)

New Features

• Support SimMIM (#239)
• Add KNN benchmark, support KNN test with checkpoint and extracted backbone weights (#243)
• Support ImageNet-21k dataset (#225)
• Resume latest checkpoint automatically (#245)

Bug Fixes

• Add seed to distributed sampler (#250)
• Fix positional parameter error in dist_test_svm_epoch.sh (#260)
• Fix 'mkdir' error in prepare_voc07_cls.sh (#261)

Improvements

• Update args format from command line (#253)

Docs

• Fix command errors in 6_benchmarks.md (#263)
• Translate 6_benchmarks.md to Chinese (#262)

Contributors

A total of 5 developers contributed to this release. Thanks @Jiahao000 , @fangyixiao18 , @YuanLiuuuuuu , @soonera , @mitming

Bug Fixes

Fix MMClassification version dependency

Highlight

• Support MAE (#221)
• Add Places205 benchmarks (#210)
• Add test Windows in workflows (#215)

New Features

• Support MAE (#221)
• Add Places205 benchmarks (#210)

Bug Fixes

• Fix config typos for rotation prediction and deepcluster (#200)
• Fix image channel bgr/rgb bug and update benchmarks (#210)
• Fix the bug when using prefetch under multi-view methods (#218)
• Fix tsne 'no init_cfg' error (#222)

Improvements

• Deprecate imgs_per_gpu and use samples_per_gpu (#204)
• Update the installation of MMCV (#208)
• Add pre-commit hook for algo-readme and copyright (#213)
• Add test Windows in workflows (#215)

Docs

• Translate 0_config.md into Chinese (#216)
• Reorganizing OpenMMLab projects and update algorithms in readme (#219)

Contributors

A total of 6 developers contributed to this release. Thanks @Jiahao000 , @fangyixiao18 , @YuanLiuuuuuu , @wang11wang , @c-liangyu , @mitming

Highlight

• Support vision transformer based MoCo v3 (#194)
• Speed up training and start time (#181)
• Support cpu training (#188)

New Features

• Support vision transformer based MoCo v3 (#194)
• Support cpu training (#188)

Bug Fixes

• Fix issue (#159, #160) related bugs (#161)
• Fix missing prob assignment in RandomAppliedTrans (#173)
• Fix bug of showing k-means losses (#182)
• Fix bug in non-distributed multi-gpu training/testing (#189)
• Fix bug when loading cifar dataset (#191)
• Fix dataset.evaluate args bug (#192)

Improvements

• Cancel previous runs that are not completed in CI (#145)
• Enhance MIM function (#152)
• Skip CI when some specific files were changed (#154)
• Add drop_last when building eval optimizer (#158)
• Deprecate the support for "python setup.py test" (#174)
• Speed up training and start time (#181)
• Upgrade isort to 5.10.1 (#184)

Docs

• Refactor the directory structure of docs (#146)
• Fix readthedocs (#148, #149, #153)
• Fix typos and dead links in some docs (#155, #180, #195)
• Update training logs and benchmark results in model zoo (#157, #165, #195)
• Update and translate some docs into Chinese (#163, #164, #165, #166, #167, #168, #169, #172, #176, #178, #179)
• Update algorithm README with the new format (#177)

Contributors

A total of 10 developers contributed to this release. Thanks @Jiahao000, @fangyixiao18, @YuanLiuuuuuu, @soonera, @Kaminyou, @mitming, @RangeKing, @xcnick, @st9007a, @Muyun99

New Contributors

• @RangeKing made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/163
• @Muyun99 made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/168
• @xcnick made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/173
• @soonera made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/178
• @Kaminyou made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/180
• @st9007a made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/184
• @mitming made their first contribution in https://github.com/open-mmlab/mmselfsup/pull/191

Highlight

• Released with code refactor.
• Add 3 new self-supervised learning algorithms.
• Support benchmarks with MMDet and MMSeg.
• Add comprehensive documents.

Refactor

• Merge redundant dataset files.
• Adapt to new version of MMCV and remove old version related codes.
• Inherit MMCV BaseModule.
• Optimize directory.
• Rename all config files.

New Features

• Add SwAV, SimSiam, DenseCL algorithms.
• Add t-SNE visualization tools.
• Support MMCV version fp16.

Benchmarks

• More benchmarking results, including classification, detection and segmentation.
• Support some new datasets in downstream tasks.
• Launch MMDet and MMSeg training with MIM.

Docs

• Refactor README, getting_started, install, model_zoo files.
• Add data_prepare file.
• Add comprehensive tutorials.