use Dynuent ,i hava a bug that :" seg_prob = predictor(window_data, *args, **kwargs).to(device) 'list' object has no attribute 'to' ”

[gaoge230]

Hello , I had a problem with dynunet，
I used covd-19 challenge. The following error occurred referring to dynunet tutorial.

The problem is described as follows：

raceback (most recent call last):
File "/zengxh_fix/gg/code/run_net_our.py", line 434, in
train(data_folder=data_folder, model_folder=args.model_folder)
File "/zengxh_fix/gg/code/run_net_our.py", line 346, in train
trainer.run()
File "/zengxh_fix/gg/download/MONAI-master/monai/engines/trainer.py", line 46, in run
super().run()
File "/zengxh_fix/gg/download/MONAI-master/monai/engines/workflow.py", line 163, in run
super().run(data=self.data_loader, max_epochs=self.state.max_epochs)
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 691, in run
return self._internal_run()
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 762, in _internal_run
self._handle_exception(e)
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 465, in _handle_exception
self._fire_event(Events.EXCEPTION_RAISED, e)
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 423, in _fire_event
func(*first, *(event_args + others), **kwargs)
File "/zengxh_fix/gg/download/MONAI-master/monai/handlers/stats_handler.py", line 145, in exception_raised
raise e
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 730, in _internal_run
time_taken = self._run_once_on_dataset()
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 828, in _run_once_on_dataset
self._handle_exception(e)
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 465, in _handle_exception
self._fire_event(Events.EXCEPTION_RAISED, e)
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 423, in _fire_event
func(*first, *(event_args + others), **kwargs)
File "/zengxh_fix/gg/download/MONAI-master/monai/handlers/stats_handler.py", line 145, in exception_raised
raise e
File "/opt/conda/lib/python3.7/site-packages/ignite/engine/engine.py", line 811, in _run_once_on_dataset
self.state.output = self._process_function(self, self.state.batch)
File "/zengxh_fix/gg/download/MONAI-master/monai/engines/trainer.py", line 153, in _iteration
predictions = self.inferer(inputs, self.network, *args, **kwargs)
File "/zengxh_fix/gg/download/MONAI-master/monai/inferers/inferer.py", line 174, in call
**kwargs,
File "/zengxh_fix/gg/download/MONAI-master/monai/inferers/utils.py", line 127, in sliding_window_inference
seg_prob = predictor(window_data, *args, **kwargs).to(device) # batched patch segmentation
AttributeError: 'list' object has no attribute 'to'

Can you answer for me？Thank you

[Nic-Ma]

Hi @gaoge230 ,

Thanks for your interest here.
Are you using this COVID-19 challenge tutorial?
https://github.com/Project-MONAI/tutorials/tree/master/3d_segmentation/challenge_baseline
But I think it uses UNet instead of DynUNet.

Thanks.

[Nic-Ma]

And I also tried to run the DynUNet tutorial which is based on Decathon dataset:
https://github.com/Project-MONAI/tutorials/blob/master/modules/dynunet_tutorial.ipynb
No issue happened.

Thanks.

[wyli]

I think this is valid issue, it could happen when deep_supervision=True in dynunet, the network forward will return a list instead of a tensor. but I'm not sure how we improve the usage here cc @yiheng-wang-nv

[gaoge230]

@wyli Thank you very much .Your answer helped me solve the problem。good!!!!! Thanks!!!!!!!!!!!!!haha

[gaoge230]

@Nic-Ma Thank you very much. The problem has been solved, using deep_supervision=False. but when deep_supervision =True ，the problem will arise.

[rijobro]

Could we have __call__ just return the image, and then have an extra method get_feature_maps, which would require deep_supervision==True. In this way, the return of net(x) would be consistent regardless of deep_supervision.