Jialian Wu

It is not a problem for not using a pretrained model or the warnings. If you have reasonable results before post-processing, it looks like not a problem from trained model....

The cuda installation command is shown in the INSTALL.md as: conda install pytorch=1.3.1 torchvision=0.4.2 cudatoolkit=10.0.130 -c pytorch

TraDeS should work for MOT20, where I saw some papers reporting TraDeS performance on it. But I didn't do this experiment.

This is a common identity-switch problem appeared in most existing methods to some extent. There is no simple solution to solve this in this project. Probably you can try to...

https://github.com/JialianW/TraDeS/blob/3eafd249ca0f18af8000d5798d4c552a0bd627ec/experiments/youtube_vis.sh

@futureisatyourhand I cite the text in our paper as your reference: "For the visual and geometrical trees, we employ K-means to cluster the visual features and ground truth binary masks...