search

alibaba / u2mot

[ICCV 2023] Uncertainty-aware Unsupervised Multi-Object Tracking
MIT License
18 stars 0 forks source link

eval result on visdrone test set when using VisDrone2018-MOT-toolkit #2

Open RenLibo-aircas opened 5 months ago

RenLibo-aircas commented 5 months ago

Hello, I really like your paper, but I have a question. Regarding the validation script on the VisDronedataset, are you using the py-motmetrics library for validation instead of the official VisDrone matlab toolkil？(https://github.com/VisDrone/VisDrone2018-MOT-toolkit) I tested your model and obtained an MOTA of 55.9, but when using the MATLAB library, the MOTA is only 50.2. It seems that there may be some differences in the validation process between the two libraries, and I haven't found the reason.

kail8 commented 1 month ago

Sorry for the late reply. I am working on LLM and MLLM now, and I am losing some details of this paper. But it seems we did use the py-motmetrics library for evaluation to align with other MOT datasets in this repo. We will update the results with the VisDrone toolkit if available.