search

JialianW / TraDeS

Track to Detect and Segment: An Online Multi-Object Tracker (CVPR 2021)
MIT License
553 stars 108 forks source link

How to try to train a model of our own（crowdhuman dataset） #35

Closed 15835111234 closed 2 years ago

15835111234 commented 3 years ago

We use the script and data set you provide to try training ；Training is ok, but an error is reported during verificationwe ：assert len(cost_volume_track_ids) == self.opt.clip_len 。

Use our own data set to also include such errors 。

JialianW commented 3 years ago

Thanks for the interests!

Did you try the testing script? What do you mean by verification? We didn't provide code for evaluating the model while training.