ocetintas/mpntrackseg

"Multi-Object Tracking and Segmentation via Neural Message Passing" (IJCV 2022) - Official Implementation

mpntrackseg's Issues

hang at init

I'm seeing the code hang when trying to train the model at this line of code: https://github.com/ocetintas/MPNTrackSeg/blob/main/src/mot_neural_solver/pl_module/pl_module.py#L42

This call (perhaps the mapping to cuda() ) doesn't seem to ever complete. Seen this issue at all?

KittiMOTS detection file does not have rle

Hi. Ocentintas
Thank you for your great work in MOTS.
I have a question about KittiMOTS detection file.
when I run the code python scripts/train.py with KittiMOTS datasets.
I got this error.
ValueError: Length mismatch: Expected axis has 9 elements, new values have 10 elements
I check Kittiloader code and I realize column has 'rle' which is related to mask but detection file which is dowloaded with yours does not have mask representations.
How can I train KITTI MOTS?

Issue with MOT17

Hi, Thank you for the code.
I have a question about running MOT17.
We ran the old code of mot_neural_solver (https://github.com/dvl-tum/mot_neural_solver) for MOT17, Where the accuracy of MOT17 was low. So we tried this code as you have used edge classification + mask prediction to improve the accuracy.
But unfortunately, we are facing issues with the MOT17 train as it gives an error AttributeError: 'NoneType' object has no attribute ‘to. pickle’.

Can we run MOT17 with the same code as MOTS20?

Kindly let us know how to proceed further.

Thank you.