Quick test failure (axondeepseg_test)

[shinysyk]

Hi,

I installed ADS by following the instructions of https://axondeepseg.readthedocs.io/en/latest/documentation.html. However, the quick test (axondeepseg_test) failed with the following error message. Of note, I am using Ubuntu 22.04.2 LTS.

Traceback (most recent call last):
File "/home/yungki/anaconda3/envs/ads_venv/bin/axondeepseg_test", line 33, in
sys.exit(load_entry_point('AxonDeepSeg', 'console_scripts', 'axondeepseg_test')())
File "/home/yungki/axondeepseg/AxonDeepSeg/integrity_test.py", line 35, in integrity_test
axon_segmentation(path_testing, [str(path_testing / image)], path_model, acquired_resolution=0.13, gpu_id=0, overlap_value=[48,48])
File "/home/yungki/axondeepseg/AxonDeepSeg/apply_model.py", line 61, in axon_segmentation
nii_lst, _ = imed_inference.segment_volume(str(path_model), input_filenames, gpu_id=gpu_id, options=options)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/ivadomed/inference.py", line 498, in segment_volume
preds = get_preds(context, fname_model, model_params, cuda_available, device, batch)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/ivadomed/inference.py", line 84, in get_preds
preds = model(img)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/ivadomed/models.py", line 509, in forward
features, _ = self.encoder(x)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/ivadomed/models.py", line 343, in forward
x = self.down_path0
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/ivadomed/models.py", line 255, in forward
x = F.relu(self.conv1(x))
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 447, in forward
return self._conv_forward(input, self.weight, self.bias)
File "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 443, in _conv_forward
return F.conv2d(input, weight, bias, self.stride,
RuntimeError: no valid convolution algorithms available in CuDNN

Any idea about how to fix this error? Thanks in advance!!!

[Stoyan-I-A]

Humm this is interesting. Thank you @shinysyk for reaching out.
It seems like Ivadomed, which is the package we use for applying a network, detected that you have a dedicated GPU and tried to do the segmentation using CUDA (which would run on the GPU). But using CUDA requires installing other stuff like a specific version of CuDNN related to the version of CUDA used.

At least this is what I think.

A workaround around this would be to open the file "/home/yungki/anaconda3/envs/ads_venv/lib/python3.8/site-packages/ivadomed/inference.py", go to line 498 and change
cuda_available to False
and
device to "cpu"

This should force the software to use the CPU instead of the GPU to do the segmentation.

I admit that this is not a proper solution and I need to discuss with the team of a better way to do this.

[shinysyk]

Thanks a lot! It worked.