In the eval_seg method in utils.py, the case where c > 2 is not working. It will prompt that pred is a numpy array and does not have a float() method. If torch.tensor(pred) is used at this point, it will cause vpred_cpu = vpred.cpu() to lose the channel dimension directly. I don't know why this happens, but when I write the code following the c == 2 case, everything works fine except that the vis export will duplicate each image 5 times. I hope to get some help.
In the eval_seg method in utils.py, the case where c > 2 is not working. It will prompt that pred is a numpy array and does not have a float() method. If torch.tensor(pred) is used at this point, it will cause vpred_cpu = vpred.cpu() to lose the channel dimension directly. I don't know why this happens, but when I write the code following the c == 2 case, everything works fine except that the vis export will duplicate each image 5 times. I hope to get some help.
