Roadmap of MMTracking

[OceanPang]

We keep this issue open to collect feature requests from users and hear your voice.

You can either:

1. Suggest a new feature by leaving a comment.
2. Vote for a feature request with 👍 or be against with 👎. (Remember that developers are busy and cannot respond to all feature requests, so vote for your most favorable one!)
3. Tell us that you would like to help implement one of the features in the list or review the PRs. (This is the greatest things to hear about!)

Recently we do not have enough bandwidth/developers to support new methods. If you are interested in joining us as an intern or a full-time researcher/engineer, feel free to let us know. You can directly drop an email to gongtao@pjlab.org.cn, pangjiangmiao@gmail.com, or chenkai@sensetime.com.

Besides the developments from OpenMMLab, we also welcome all contributions from the community. You can make a PR of your work to this repository following CONTRIBUTING.md.

[SilvioGiancola]

Thank you for the amazing work! Do you have any plan on releasing more baselines on Single Object Tracking (Siam-like, ATOM, DiMP, OCEAN, KYS, LWL) as well as more dataset (OTB, VOT, TrackingNet, UAV123, NFS)?

[agirbau]

Hello! I think CenterTrack would be a great incorporation to the multiple object tracking.

[OceanPang]

Hi @SilvioGiancola @agirbau, thanks for your interest!

We are discussing a development plan in which we tend to support 2-3 new methods for each task and support the other popular datasets.

For the methods, the Siam-Like and some recent SOTA methods (not decided which one yet) are the first priority for SOT @SilvioGiancola. As @agirbau suggested, We do have a plan to support CenterTrack but need to wait for the CenterNet in mmdetection.

For the datasets, we will support the datasets @SilvioGiancola mentioned, and also support some other large scale MOT benchmarks.

However, considering the limited bandwidth, we cannot clearly state the release schedule of these updates. If you are also interested in supporting these features, we will really appreciate your pull requests. : )

[imr555]

Just wanted to add, it would also be cool, If a multi object tracking and segmentation baseline could be added like
(TrackRCNN)
https://github.com/VisualComputingInstitute/TrackR-CNN
for example.
As many benchmarks and workshops are introducing datasets and problems specifically in this domain such as MOTS CHALLENGE, KITTI MOTS, BDD100K MOTS etc.

And again, really appreciate this amazing work, for putting everything in such a modular perspective with efficiency.

[shwoo93]

Thanks for your hard work on maintaining this wonderful repo.
It would be nice to include recently released large-vocabulary tracking data TAO as well.

[RaymondByc]

When will release the CenterTrack in your plan?

[dzambrano]

What about HOTA metrics?

[austinmw]

Hi, I'd like to suggest StrongSORT MOT algorithm #541

[noahcao]

I am considering to make support to some features in the short future. Let me know if some are already in process or your vote for their priority.

1. Support of DanceTrack[0] dataset.( [Feature] Support DanceTrack dataset for MOT #543)
2. Support of OC-SORT[6] for multi-object tracking. ([Feature] Support OC-SORT for MOT #545 )
3. multi-class multi-object tracking, which is already supported in Youtube-VIS but still absent on MOT task. ([Feature] Support multi-class tracking in ByteTrack #548)
4. transformer-based detector, e.g. DETR[1], and tracker, e.g. TransTrack[2], Trackformer[3].
5. more MOT datasets, such as KITTI[4].
6. make BYTE[5] and OC-SORT optional flags in other KalmanFilter-based tracking methods.

[0] "DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion"
[1] "End-to-end object detection with transformers"
[2] "Transtrack: Multiple object tracking with transformer"
[3] "Trackformer: Multi-object tracking with transformers"
[4] "Vision meets robotics: The kitti dataset"
[5] "ByteTrack: Multi-Object Tracking by Associating Every Detection Box"
[6] "Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking"

[noahcao]

Pose track is also a good to-do as a new task, baseline methods include PoseFlow[1]. The corresponding datasets are PoseTrack[2] and HiEve[3].

[1] "Pose Flow: Efficient online pose tracking"
[2] "Posetrack: A benchmark for human pose estimation and tracking"
[3] "Human in events: A large-scale benchmark for human-centric video analysis in complex events"

An existing codebase Alphapose may be a good reference for the practice, which focuses on pose estimation while leaving tracking modules not actively maintained.