Skip to content

Latest commit

 

History

History
612 lines (423 loc) · 31.7 KB

changelog.md

File metadata and controls

612 lines (423 loc) · 31.7 KB
Raw

Changelog

V0.22.1 (3/9/2022)

Bug Fixes

  • Fix the ZeroDivisionError that all pixels in one image is ignored. (#1336)

Improvements

  • Provide URLs of STDC, Segmenter and Twins pretrained models (#1272)

V0.22 (3/04/2022)

Highlights

  • Support ConvNeXt: A ConvNet for the 2020s. Please use the latest MMClassification (0.21.0) to try it out.
  • Support iSAID aerial Dataset.
  • Officially Support inference on Windows OS.

New Features

  • Support ConvNeXt: A ConvNet for the 2020s. (#1216)
  • Support iSAID aerial Dataset. (#1115
  • Generating and plotting confusion matrix. (#1301)

Improvements

  • Refactor 4 decoder heads (ASPP, FCN, PSP, UPer): Split forward function into _forward_feature and cls_seg. (#1299)
  • Add min_size arg in Resize to keep the shape after resize bigger than slide window. (#1318)
  • Revise pre-commit-hooks. (#1315)
  • Add win-ci. (#1296)

Bug Fixes

  • Fix mlp_ratio type in Swin Transformer. (#1274)
  • Fix path errors in ./demo . (#1269)
  • Fix bug in conversion of potsdam. (#1279)
  • Make accuracy take into account ignore_index. (#1259)
  • Add Pytorch HardSwish assertion in unit test. (#1294)
  • Fix wrong palette value in vaihingen. (#1292)
  • Fix the bug that SETR cannot load pretrain. (#1293)
  • Update correct In Collection in metafile of each configs. (#1239)
  • Upload completed STDC models. (#1332)
  • Fix DNLHead exports onnx inference difference type Cast error. (#1161)

Contributors

V0.21.1 (2/9/2022)

Bug Fixes

  • Fix typos in docs. (#1263)
  • Fix repeating log by setup_multi_processes. (#1267)
  • Upgrade isort in pre-commit hook. (#1270)

Improvements

  • Use MMCV load_state_dict func in ViT/Swin. (#1272)
  • Add exception for PointRend for support CPU-only. (#1271)

V0.21 (1/29/2022)

Highlights

  • Officially Support CPUs training and inference, please use the latest MMCV (1.4.4) to try it out.
  • Support Segmenter: Transformer for Semantic Segmentation (ICCV'2021).
  • Support ISPRS Potsdam and Vaihingen Dataset.
  • Add Mosaic transform and MultiImageMixDataset class in dataset_wrappers.

New Features

  • Support Segmenter: Transformer for Semantic Segmentation (ICCV'2021) (#955)
  • Support ISPRS Potsdam and Vaihingen Dataset (#1097, #1171)
  • Add segformer‘s benchmark on cityscapes (#1155)
  • Add auto resume (#1172)
  • Add Mosaic transform and MultiImageMixDataset class in dataset_wrappers (#1093, #1105)
  • Add log collector (#1175)

Improvements

  • New-style CPU training and inference (#1251)
  • Add UNet benchmark with multiple losses supervision (#1143)

Bug Fixes

  • Fix the model statistics in doc for readthedoc (#1153)
  • Set random seed for palette if not given (#1152)
  • Add COCOStuffDataset in class_names.py (#1222)
  • Fix bug in non-distributed multi-gpu training/testing (#1247)
  • Delete unnecessary lines of STDCHead (#1231)

Contributors

V0.20.2 (12/15/2021)

Bug Fixes

  • Revise --option to --options to avoid BC-breaking. (#1140)

V0.20.1 (12/14/2021)

Improvements

  • Change options to cfg-options (#1129)

Bug Fixes

  • Fix <!-- [ABSTRACT] --> in metafile. (#1127)
  • Fix correct num_classes of HRNet in LoveDA dataset (#1136)

V0.20 (12/10/2021)

Highlights

  • Support Twins (#989)
  • Support a real-time segmentation model STDC (#995)
  • Support a widely-used segmentation model in lane detection ERFNet (#960)
  • Support A Remote Sensing Land-Cover Dataset LoveDA (#1028)
  • Support focal loss (#1024)

New Features

  • Support Twins (#989)
  • Support a real-time segmentation model STDC (#995)
  • Support a widely-used segmentation model in lane detection ERFNet (#960)
  • Add SETR cityscapes benchmark (#1087)
  • Add BiSeNetV1 COCO-Stuff 164k benchmark (#1019)
  • Support focal loss (#1024)
  • Add Cutout transform (#1022)

Improvements

  • Set a random seed when the user does not set a seed (#1039)
  • Add CircleCI setup (#1086)
  • Skip CI on ignoring given paths (#1078)
  • Add abstract and image for every paper (#1060)
  • Create a symbolic link on windows (#1090)
  • Support video demo using trained model (#1014)

Bug Fixes

  • Fix incorrectly loading init_cfg or pretrained models of several transformer models (#999, #1069, #1102)
  • Fix EfficientMultiheadAttention in SegFormer (#1037)
  • Remove fp16 folder in configs (#1031)
  • Fix several typos in .yml file (Dice Metric #1041, ADE20K dataset #1120, Training Memory (GB) #1083)
  • Fix test error when using --show-dir (#1091)
  • Fix dist training infinite waiting issue (#1035)
  • Change the upper version of mmcv to 1.5.0 (#1096)
  • Fix symlink failure on Windows (#1038)
  • Cancel previous runs that are not completed (#1118)
  • Unified links of readthedocs in docs (#1119)

Contributors

V0.19 (11/02/2021)

Highlights

  • Support TIMMBackbone wrapper (#998)
  • Support custom hook (#428)
  • Add codespell pre-commit hook (#920)
  • Add FastFCN benchmark on ADE20K (#972)

New Features

  • Support TIMMBackbone wrapper (#998)
  • Support custom hook (#428)
  • Add FastFCN benchmark on ADE20K (#972)
  • Add codespell pre-commit hook and fix typos (#920)

Improvements

  • Make inputs & channels smaller in unittests (#1004)
  • Change self.loss_decode back to dict in Single Loss situation (#1002)

Bug Fixes

  • Fix typo in usage example (#1003)
  • Add contiguous after permutation in ViT (#992)
  • Fix the invalid link (#985)
  • Fix bug in CI with python 3.9 (#994)
  • Fix bug when loading class name form file in custom dataset (#923)

Contributors

V0.18 (10/07/2021)

Highlights

  • Support three real-time segmentation models (ICNet #884, BiSeNetV1 #851, and BiSeNetV2 #804)
  • Support one efficient segmentation model (FastFCN #885)
  • Support one efficient non-local/self-attention based segmentation model (ISANet #70)
  • Support COCO-Stuff 10k and 164k datasets (#625)
  • Support evaluate concated dataset separately (#833)
  • Support loading GT for evaluation from multi-file backend (#867)

New Features

  • Support three real-time segmentation models (ICNet #884, BiSeNetV1 #851, and BiSeNetV2 #804)
  • Support one efficient segmentation model (FastFCN #885)
  • Support one efficient non-local/self-attention based segmentation model (ISANet #70)
  • Support COCO-Stuff 10k and 164k datasets (#625)
  • Support evaluate concated dataset separately (#833)

Improvements

  • Support loading GT for evaluation from multi-file backend (#867)
  • Auto-convert SyncBN to BN when training on DP automatly(#772)
  • Refactor Swin-Transformer (#800)

Bug Fixes

  • Update mmcv installation in dockerfile (#860)
  • Fix number of iteration bug when resuming checkpoint in distributed train (#866)
  • Fix parsing parse in val_step (#906)

V0.17 (09/01/2021)

Highlights

  • Support SegFormer
  • Support DPT
  • Support Dark Zurich and Nighttime Driving datasets
  • Support progressive evaluation

New Features

  • Support SegFormer (#599)
  • Support DPT (#605)
  • Support Dark Zurich and Nighttime Driving datasets (#815)
  • Support progressive evaluation (#709)

Improvements

  • Add multiscale_output interface and unittests for HRNet (#830)
  • Support inherit cityscapes dataset (#750)
  • Fix some typos in README.md (#824)
  • Delete convert function and add instruction to ViT/Swin README.md (#791)
  • Add vit/swin/mit convert weight scripts (#783)
  • Add copyright files (#796)

Bug Fixes

  • Fix invalid checkpoint link in inference_demo.ipynb (#814)
  • Ensure that items in dataset have the same order across multi machine (#780)
  • Fix the log error (#766)

V0.16 (08/04/2021)

Highlights

  • Support PyTorch 1.9
  • Support SegFormer backbone MiT
  • Support md2yml pre-commit hook
  • Support frozen stage for HRNet

New Features

  • Support SegFormer backbone MiT (#594)
  • Support md2yml pre-commit hook (#732)
  • Support mim (#717)
  • Add mmseg2torchserve tool (#552)

Improvements

  • Support hrnet frozen stage (#743)
  • Add template of reimplementation questions (#741)
  • Output pdf and epub formats for readthedocs (#742)
  • Refine the docstring of ResNet (#723)
  • Replace interpolate with resize (#731)
  • Update resource limit (#700)
  • Update config.md (#678)

Bug Fixes

  • Fix ATTENTION registry (#729)
  • Fix analyze log script (#716)
  • Fix doc api display (#725)
  • Fix patch_embed and pos_embed mismatch error (#685)
  • Fix efficient test for multi-node (#707)
  • Fix init_cfg in resnet backbone (#697)
  • Fix efficient test bug (#702)
  • Fix url error in config docs (#680)
  • Fix mmcv installation (#676)
  • Fix torch version (#670)

Contributors

@sshuair @xiexinch @Junjun2016 @mmeendez8 @xvjiarui @sennnnn @puhsu @BIGWangYuDong @keke1u @daavoo

V0.15 (07/04/2021)

Highlights

  • Support ViT, SETR, and Swin-Transformer
  • Add Chinese documentation
  • Unified parameter initialization

Bug Fixes

  • Fix typo and links (#608)
  • Fix Dockerfile (#607)
  • Fix ViT init (#609)
  • Fix mmcv version compatible table (#658)
  • Fix model links of DMNEt (#660)

New Features

  • Support loading DeiT weights (#538)
  • Support SETR (#531, #635)
  • Add config and models for ViT backbone with UperHead (#520, #635)
  • Support Swin-Transformer (#511)
  • Add higher accuracy FastSCNN (#606)
  • Add Chinese documentation (#666)

Improvements

  • Unified parameter initialization (#567)
  • Separate CUDA and CPU in github action CI (#602)
  • Support persistent dataloader worker (#646)
  • Update meta file fields (#661, #664)

V0.14 (06/02/2021)

Highlights

  • Support ONNX to TensorRT
  • Support MIM

Bug Fixes

  • Fix ONNX to TensorRT verify (#547)
  • Fix save best for EvalHook (#575)

New Features

  • Support loading DeiT weights (#538)
  • Support ONNX to TensorRT (#542)
  • Support output results for ADE20k (#544)
  • Support MIM (#549)

Improvements

  • Add option for ViT output shape (#530)
  • Infer batch size using len(result) (#532)
  • Add compatible table between MMSeg and MMCV (#558)

V0.13 (05/05/2021)

Highlights

  • Support Pascal Context Class-59 dataset.
  • Support Visual Transformer Backbone.
  • Support mFscore metric.

Bug Fixes

  • Fixed Colaboratory tutorial (#451)
  • Fixed mIoU calculation range (#471)
  • Fixed sem_fpn, unet README.md (#492)
  • Fixed num_classes in FCN for Pascal Context 60-class dataset (#488)
  • Fixed FP16 inference (#497)

New Features

  • Support dynamic export and visualize to pytorch2onnx (#463)
  • Support export to torchscript (#469, #499)
  • Support Pascal Context Class-59 dataset (#459)
  • Support Visual Transformer backbone (#465)
  • Support UpSample Neck (#512)
  • Support mFscore metric (#509)

Improvements

  • Add more CI for PyTorch (#460)
  • Add print model graph args for tools/print_config.py (#451)
  • Add cfg links in modelzoo README.md (#468)
  • Add BaseSegmentor import to segmentors/init.py (#495)
  • Add MMOCR, MMGeneration links (#501, #506)
  • Add Chinese QR code (#506)
  • Use MMCV MODEL_REGISTRY (#515)
  • Add ONNX testing tools (#498)
  • Replace data_dict calling 'img' key to support MMDet3D (#514)
  • Support reading class_weight from file in loss function (#513)
  • Make tags as comment (#505)
  • Use MMCV EvalHook (#438)

V0.12 (04/03/2021)

Highlights

  • Support FCN-Dilate 6 model.
  • Support Dice Loss.

Bug Fixes

  • Fixed PhotoMetricDistortion Doc (#388)
  • Fixed install scripts (#399)
  • Fixed Dice Loss multi-class (#417)

New Features

  • Support Dice Loss (#396)
  • Add plot logs tool (#426)
  • Add opacity option to show_result (#425)
  • Speed up mIoU metric (#430)

Improvements

  • Refactor unittest file structure (#440)
  • Fix typos in the repo (#449)
  • Include class-level metrics in the log (#445)

V0.11 (02/02/2021)

Highlights

  • Support memory efficient test, add more UNet models.

Bug Fixes

  • Fixed TTA resize scale (#334)
  • Fixed CI for pip 20.3 (#307)
  • Fixed ADE20k test (#359)

New Features

  • Support memory efficient test (#330)
  • Add more UNet benchmarks (#324)
  • Support Lovasz Loss (#351)

Improvements

  • Move train_cfg/test_cfg inside model (#341)

V0.10 (01/01/2021)

Highlights

  • Support MobileNetV3, DMNet, APCNet. Add models of ResNet18V1b, ResNet18V1c, ResNet50V1b.

Bug Fixes

  • Fixed CPU TTA (#276)
  • Fixed CI for pip 20.3 (#307)

New Features

  • Add ResNet18V1b, ResNet18V1c, ResNet50V1b, ResNet101V1b models (#316)
  • Support MobileNetV3 (#268)
  • Add 4 retinal vessel segmentation benchmark (#315)
  • Support DMNet (#313)
  • Support APCNet (#299)

Improvements

  • Refactor Documentation page (#311)
  • Support resize data augmentation according to original image size (#291)

V0.9 (30/11/2020)

Highlights

  • Support 4 medical dataset, UNet and CGNet.

New Features

  • Support RandomRotate transform (#215, #260)
  • Support RGB2Gray transform (#227)
  • Support Rerange transform (#228)
  • Support ignore_index for BCE loss (#210)
  • Add modelzoo statistics (#263)
  • Support Dice evaluation metric (#225)
  • Support Adjust Gamma transform (#232)
  • Support CLAHE transform (#229)

Bug Fixes

  • Fixed detail API link (#267)

V0.8 (03/11/2020)

Highlights

  • Support 4 medical dataset, UNet and CGNet.

New Features

  • Support customize runner (#118)
  • Support UNet (#161)
  • Support CHASE_DB1, DRIVE, STARE, HRD (#203)
  • Support CGNet (#223)

V0.7 (07/10/2020)

Highlights

  • Support Pascal Context dataset and customizing class dataset.

Bug Fixes

  • Fixed CPU inference (#153)

New Features

  • Add DeepLab OS16 models (#154)
  • Support Pascal Context dataset (#133)
  • Support customizing dataset classes (#71)
  • Support customizing dataset palette (#157)

Improvements

  • Support 4D tensor output in ONNX (#150)
  • Remove redundancies in ONNX export (#160)
  • Migrate to MMCV DepthwiseSeparableConv (#158)
  • Migrate to MMCV collect_env (#137)
  • Use img_prefix and seg_prefix for loading (#153)

V0.6 (10/09/2020)

Highlights

  • Support new methods i.e. MobileNetV2, EMANet, DNL, PointRend, Semantic FPN, Fast-SCNN, ResNeSt.

Bug Fixes

  • Fixed sliding inference ONNX export (#90)

New Features

  • Support MobileNet v2 (#86)
  • Support EMANet (#34)
  • Support DNL (#37)
  • Support PointRend (#109)
  • Support Semantic FPN (#94)
  • Support Fast-SCNN (#58)
  • Support ResNeSt backbone (#47)
  • Support ONNX export (experimental) (#12)

Improvements

  • Support Upsample in ONNX (#100)
  • Support Windows install (experimental) (#75)
  • Add more OCRNet results (#20)
  • Add PyTorch 1.6 CI (#64)
  • Get version and githash automatically (#55)

v0.5.1 (11/08/2020)

Highlights

  • Support FP16 and more generalized OHEM

Bug Fixes

  • Fixed Pascal VOC conversion script (#19)
  • Fixed OHEM weight assign bug (#54)
  • Fixed palette type when palette is not given (#27)

New Features

  • Support FP16 (#21)
  • Generalized OHEM (#54)

Improvements

  • Add load-from flag (#33)
  • Fixed training tricks doc about different learning rates of model (#26)