Training CLRNet with custom dataset in Tulane format, get cuda device-side assert error

[aeroumar]

Hi @MySuperSoul @Turoad , thank you for this implementation! I have used the tools/detect.py script from your other lanedet repo to get great performance from the pretrained weights of culane_r101.pth for this CLRNet model.

I am now trying to train the CLRNet model using my own dataset which I have converted to the Tusimple format. I copied the clr_resnet101_tusimple.py config and tusimple.py dataset loader then modified them to point to my own data directories. However when I try to run the training using main.py, I am faced with the following error:

I appreciate any guidance you may have as to what could be causing this error.

[zillur-av]

This error shows up when the number of class in ground truth does not match the number of class in prediction.