search

Zhongdao / UniTrack

[NeurIPS'21] Unified tracking framework with a single appearance model. It supports Single Object Tracking (SOT), Video Object Segmentation (VOS), Multi-Object Tracking (MOT), Multi-Object Tracking and Segmentation (MOTS), Pose Tracking, Video Instance Segmentation (VIS), and class-agnostic MOT (e.g. TAO dataset).
MIT License
334 stars 35 forks source link

One Potential Bug #24

Closed jimmy-dq closed 2 years ago

jimmy-dq commented 2 years ago

Hi Zhongdao, https://github.com/Zhongdao/UniTrack/blob/e18aece2a9046225ec3b3dd595a35490d594c699/test/test_mot.py#L37 Thanks for your great work to provide such a wonderful evaluation framework! Recently I have used your framework to test the model on MOT16 (test set). I find when the detection result of a specific frame provided by FairMOT is empty, the 'obs' shown in the above will be also empty and the program will crash.

Maybe we can re-create a zero matrix obs (1x5) when it is empty (len(obs) = 0). Is it right?

In addition, does your current code provide the VIS evaluation? Since there is no any configurations in your yaml files.

Zhongdao commented 2 years ago

Thanks a lot @jimmy-dq ！ In the newest commit, I have solved this bug by checking the size of 'obs'.

Regarding the VIS evaluation: Yes VIS evaluation is provided, see this script. Evaluation on VIS is a bit tricky though :)