about the cls_num

[KDgggg]

hi,i'm training on my custom segmantic segmentation dataset. I have 17 classes [background, obj1,.....oj16]，i want the background count into loss.
The args.cls_num is set to 17 or 16? In my dataset annotation, the background's label is 0.

[KDgggg]

hi,i'm training on my custom segmantic segmentation dataset. I have 17 classes [background, obj1,.....oj16]，i want the background count into loss. The args.cls_num is set to 17 or 16? In my dataset annotation, the background's label is 0.

or 18?

[haoyudong-97]

Also note that since the output will go through argmax during the evaluation, i.e.,
dsc_batch = dice_coeff_multi_class(pred.argmax(dim=1).cpu(), torch.squeeze(msks.long(),1).cpu().long(),args.num_cls)
the output will be of size [B,H,W] where 0=background, 1=obj1, ..., 16=obj16.

[KDgggg]

Also note that since the output will go through argmax during the evaluation, i.e., dsc_batch = dice_coeff_multi_class(pred.argmax(dim=1).cpu(), torch.squeeze(msks.long(),1).cpu().long(),args.num_cls) the output will be of size [B,H,W] where 0=background, 1=obj1, ..., 16=obj16.

hello,i also meet the problem :Traceback (most recent call last):
train_loss += loss.item()
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

even if i set : export CUDA_VISIBLE_DEVICES=0, it dosen't work

[haoyudong-97]

Can you follow the Pytorch's suggestion by "passing CUDA_LAUNCH_BLOCKING=1" and show where the error occurs?

Usually this will be an index error, for example using 1 as background instead of 0