MMPose V0.26.0 Release

Highlights

• Support RLE (Residual Log-likelihood Estimation), ICCV'2021 (#1259) @Indigo6, @ly015
• Support Swin Transformer, ICCV'2021 (#1300) @yumendecc, @ly015
• Support PVT, ICCV'2021 and PVTv2, CVMJ'2022 (#1343) @zengwang430521
• Speed up inference and reduce CPU usage by optimizing the pre-processing pipeline (#1320) @chenxinfeng4, @liqikai9

New Features

• Support RLE (Residual Log-likelihood Estimation), ICCV'2021 (#1259) @Indigo6, @ly015
• Support Swin Transformer, ICCV'2021 (#1300) @yumendecc, @ly015
• Support PVT, ICCV'2021 and PVTv2, CVMJ'2022 (#1343) @zengwang430521
• Support FPN, CVPR'2017 (#1300) @yumendecc, @ly015

Improvements

• Speed up inference and reduce CPU usage by optimizing the pre-processing pipeline (#1320) @chenxinfeng4, @liqikai9
• Video demo supports models that requires multi-frame inputs (#1300) @liqikai9, @jin-s13
• Update benchmark regression list (#1328) @ly015, @liqikai9
• Remove unnecessary warnings in TopDownPoseTrack18VideoDataset (#1335) @liqikai9
• Improve documentation quality (#1313, #1305) @Ben-Louis, @ly015
• Update deprecating settings in configs (#1317) @ly015

Bug Fixes

• Fix a bug in human skeleton grouping that may skip the matching process unexpectedly when ignore_to_much is True (#1341) @daixinghome
• Fix a GPG key error that leads to CI failure (#1354) @ly015
• Fix bugs in distributed training script (#1338, #1298) @ly015
• Fix an upstream bug in xtoccotools that causes incorrect AP(M) results (#1308) @jin-s13, @ly015
• Fix indentiation errors in the colab tutorial (#1298) @YuanZi1501040205
• Fix incompatible model weight initialization with other OpenMMLab codebases (#1329) @274869388
• Fix HRNet FP16 checkpoints download URL (#1309) @yinaoxiong
• Fix typos in body3d_two_stage_video_demo.py (#1295) @mucozcan

Breaking Changes

• Refactor bbox processing in datasets and pipelines (#1311) @ly015, @Ben-Louis
  The bbox format conversion (xywh to center-scale) and random translation are moved from the dataset to the pipeline. The comparison between new and old version is as below:

	v0.26.0	v0.25.0
Dataset (e.g. TopDownCOCODataset)	... # Data sample only contains bbox rec.append({ 'bbox': obj['clean_bbox][:4], ... })	... # Convert bbox from xywh to center-scale center, scale = self._xywh2cs(*obj['clean_bbox'][:4]) # Data sample contains center and scale rec.append({ 'bbox': obj['clean_bbox][:4], 'center': center, 'scale': scale, ... })
Pipeline Config (e.g. HRNet+COCO)	... train_pipeline = [ dict(type='LoadImageFromFile'), # Convert bbox from xywh to center-scale dict(type='TopDownGetBboxCenterScale', padding=1.25), # Randomly shift bbox center dict(type='TopDownRandomShiftBboxCenter', shift_factor=0.16, prob=0.3), ... ]	... train_pipeline = [ dict(type='LoadImageFromFile'), ... ]
Advantage	Simpler data sample content Flexible bbox format conversion and augmentation Apply bbox random translation every epoch (instead of only applying once at the annotation loading)	-
BC Breaking	The method _xywh2cs of dataset base classes (e.g. Kpt2dSviewRgbImgTopDownDataset) will be deprecated in the future. Custom datasets will need modifications to move the bbox format conversion to pipelines.	-