Train error

[StevenAZy]

Traceback (most recent call last):
File "train.py", line 93, in
trainer[dataset_name](args, net, snapshot_path)
File "/home/lichangyong/Code/TransUNet/trainer.py", line 57, in trainer_synapse
loss_dice = dice_loss(outputs, label_batch, softmax=True)
File "/home/lichangyong/.pyenv/versions/pytorch_gpu/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/lichangyong/Code/TransUNet/utils.py", line 43, in forward
class_wise_dice.append(1.0 - dice.item())
RuntimeError: CUDA error: device-side assert triggered

[andife]

Hi,
I think I had a similar error message. The reason for this was the names, or the numerical values of my labels.
Maybe look facebookresearch/SymbolicMathematics#2
You should find several github issues looking for the expression "RuntimeError: CUDA error: device-side assert triggered".

[SM-93]

I had a similar problem, I was training the model with 13 classes instead of 9 (8 organs + background) so I had to change the num_classes in the dataset_config (train.py) to 14. Maybe you are also training with 13 classes.

[AllenPiaoHong]

I had a similar problem, I was training the model with 13 classes instead of 9 (8 organs + background) so I had to change the num_classes in the dataset_config (train.py) to 14. Maybe you are also training with 13 classes.

Hi, have you fixed the problem now. I also change the num_classes in train.py. But it doesn't work.

[StevenAZy]

I had a similar problem, I was training the model with 13 classes instead of 9 (8 organs + background) so I had to change the num_classes in the dataset_config (train.py) to 14. Maybe you are also training with 13 classes.

Hi, have you fixed the problem now. I also change the num_classes in train.py. But it doesn't work.

I changed the num_classes value to 14. The code can run successfully.

[rameses666]

I think the reason is the value of your label.Check that.

[JAYCHOU2020]

I had a similar problem, I was training the model with 13 classes instead of 9 (8 organs + background) so I had to change the num_classes in the dataset_config (train.py) to 14. Maybe you are also training with 13 classes.

Hi, have you fixed the problem now. I also change the num_classes in train.py. But it doesn't work.

I changed the num_classes value to 14. The code can run successfully.

my dataset class is 1 so I set numclass=2 but it also report the error ,even I set numclass=3 it not work

[buhuiya]

有人解决了吗