Mmtrack

New Features

- Support CPU training ([404](https://github.com/open-mmlab/mmtracking/pull/404))

Improvements

- Refactor SOT datasets ([401](https://github.com/open-mmlab/mmtracking/pull/401)), ([#402](https://github.com/open-mmlab/mmtracking/pull/402)), ([#393](https://github.com/open-mmlab/mmtracking/pull/393))

Highlights

- Support arXiv 2021 manuscript 'ByteTrack: Multi-Object Tracking by Associating Every Detection Box' ([385](https://github.com/open-mmlab/mmtracking/pull/385)), ([#383](https://github.com/open-mmlab/mmtracking/pull/383)), ([#372](https://github.com/open-mmlab/mmtracking/pull/372))
- Support ICCV 2019 paper 'Video Instance Segmentation' ([304](https://github.com/open-mmlab/mmtracking/pull/304)), ([#303](https://github.com/open-mmlab/mmtracking/pull/303)), ([#298](https://github.com/open-mmlab/mmtracking/pull/298)), ([#292](https://github.com/open-mmlab/mmtracking/pull/292))

New Features

- Support CrowdHuman dataset for MOT ([366](https://github.com/open-mmlab/mmtracking/pull/366))
- Support VOT2018 dataset for SOT ([305](https://github.com/open-mmlab/mmtracking/pull/305))
- Support YouTube-VIS dataset for VIS ([290](https://github.com/open-mmlab/mmtracking/pull/290))

Bug Fixes

- Fix two significant bugs in SOT and provide new SOT pretrained models ([349](https://github.com/open-mmlab/mmtracking/pull/349))

Improvements

- Refactor LaSOT, TrackingNet dataset and support GOT-10K datasets ([296](https://github.com/open-mmlab/mmtracking/pull/296))
- Support persisitent workers ([348](https://github.com/open-mmlab/mmtracking/pull/348))

New Features

- Support OTB100 dataset in SOT ([271](https://github.com/open-mmlab/mmtracking/pull/271))
- Support TrackingNet dataset in SOT ([268](https://github.com/open-mmlab/mmtracking/pull/268))
- Support UAV123 dataset in SOT ([260](https://github.com/open-mmlab/mmtracking/pull/260))

Bug Fixes

- Fix a bug in mot_param_search.py ([270](https://github.com/open-mmlab/mmtracking/pull/270))

Improvements

- Use PyTorch sphinx theme ([274](https://github.com/open-mmlab/mmtracking/pull/274))
- Use pycocotools instead of mmpycocotools ([263](https://github.com/open-mmlab/mmtracking/pull/263))

Highlights

- Release code of AAAI 2021 paper 'Temporal ROI Align for Video Object Recognition' ([247](https://github.com/open-mmlab/mmtracking/pull/247))
- Refactor English documentations ([243](https://github.com/open-mmlab/mmtracking/pull/243))
- Add Chinese documentations ([248](https://github.com/open-mmlab/mmtracking/pull/248)), ([#250](https://github.com/open-mmlab/mmtracking/pull/250))

New Features

- Support fp16 training and testing ([230](https://github.com/open-mmlab/mmtracking/pull/230))
- Release model using ResNeXt-101 as backbone for all VID methods ([254](https://github.com/open-mmlab/mmtracking/pull/254))
- Support the results of Tracktor on MOT15, MOT16 and MOT20 datasets ([217](https://github.com/open-mmlab/mmtracking/pull/217))
- Support visualization for single gpu test ([216](https://github.com/open-mmlab/mmtracking/pull/216))

Bug Fixes

- Fix a bug in MOTP evaluation ([235](https://github.com/open-mmlab/mmtracking/pull/235))
- Fix two bugs in reid training and testing ([249](https://github.com/open-mmlab/mmtracking/pull/249))

Improvements

- Refactor anchor in SiameseRPN++ ([229](https://github.com/open-mmlab/mmtracking/pull/229))
- Unify model initialization ([235](https://github.com/open-mmlab/mmtracking/pull/235))
- Refactor unittest ([231](https://github.com/open-mmlab/mmtracking/pull/231))

Highlights

- Fix training bugs of all three tasks ([219](https://github.com/open-mmlab/mmtracking/pull/219)), ([#221](https://github.com/open-mmlab/mmtracking/pull/221))

New Features

- Support error visualization for mot task ([212](https://github.com/open-mmlab/mmtracking/pull/212))

Bug Fixes

- Fix a bug in SOT demo ([213](https://github.com/open-mmlab/mmtracking/pull/213))

Improvements

- Use MMCV registry ([220](https://github.com/open-mmlab/mmtracking/pull/220))
- Add README.md for reid training ([210](https://github.com/open-mmlab/mmtracking/pull/210))
- Modify dict keys of the outputs of SOT ([223](https://github.com/open-mmlab/mmtracking/pull/223))
- Add Chinese docs including install.md, quick_run.md, model_zoo.md, dataset.md ([205](https://github.com/open-mmlab/mmtracking/pull/205)), ([#214](https://github.com/open-mmlab/mmtracking/pull/214))

New Features

- Support ReID training ([177](https://github.com/open-mmlab/mmtracking/pull/177)), ([#179](https://github.com/open-mmlab/mmtracking/pull/179)), ([#180](https://github.com/open-mmlab/mmtracking/pull/180)), ([#181](https://github.com/open-mmlab/mmtracking/pull/181)),
- Support MIM ([158](https://github.com/open-mmlab/mmtracking/pull/158))

Bug Fixes

- Fix evaluation hook ([176](https://github.com/open-mmlab/mmtracking/pull/176))
- Fix a typo in vid config ([171](https://github.com/open-mmlab/mmtracking/pull/171))

Improvements

- Refactor nms config ([167](https://github.com/open-mmlab/mmtracking/pull/167))