imagirom / confnets Goto Github PK

import confnets
model = confnets.UNet3d(depth=3, in_channels=1, out_channels=1, fmaps=[5,5,5,5])

Gave me the following error:

---

ValueError Traceback (most recent call last)
in
----> 1 model = confnets.UNet3d(depth=3, in_channels=1, out_channels=1, fmaps=[5,5,5,5])

~/anaconda3/lib/python3.7/site-packages/confnets/models/unet.py in init(self, conv_type, **super_kwargs)
285 def init(self, conv_type='Conv3d', **super_kwargs):
286 super_kwargs['conv_type'] = conv_type
--> 287 super(UNet3d, self).init(dim=3, **super_kwargs)
288
289

~/anaconda3/lib/python3.7/site-packages/confnets/models/unet.py in init(self, dim, scale_factor, conv_type, norm_type, activation, final_activation, upsampling_mode, skip_factor, *super_args, **super_kwargs)
224 with torch.no_grad():
225 inp = torch.zeros((2, self.in_channels, *self.divisibility_constraint), dtype=torch.float32)
--> 226 self(inp)
227
228 # delete attributes that are only relevant for construction and might lead to errors when model is saved

~/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py in call(self, *input, **kwargs)
530 result = self._slow_forward(*input, **kwargs)
531 else:
--> 532 result = self.forward(*input, **kwargs)
533 for hook in self._forward_hooks.values():
534 hook_result = hook(self, input, result)

~/anaconda3/lib/python3.7/site-packages/confnets/models/unet.py in forward(self, input_)
273 f'Input shape {input_.shape[2:]} not suited for downsampling with factors {self.scale_factors}.'
274 f'Lengths of spatial axes must be multiples of {self.divisibility_constraint}.'
--> 275 return super(UNet, self).forward(input_)
276
277

~/anaconda3/lib/python3.7/site-packages/confnets/models/unet.py in forward(self, input)
48 for encoded_state, upsample, skip, merge, decode in reversed(list(zip(
49 encoded_states, self.upsampling_modules, self.skip_modules, self.merge_modules, self.decoder_modules))):
---> 50 current = upsample(current)
51 encoded_state = skip(encoded_state)
52 current = merge(current, encoded_state)

~/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py in call(self, *input, **kwargs)
530 result = self._slow_forward(*input, **kwargs)
531 else:
--> 532 result = self.forward(*input, **kwargs)
533 for hook in self._forward_hooks.values():
534 hook_result = hook(self, input, result)

~/anaconda3/lib/python3.7/site-packages/confnets/layers/basic.py in forward(self, input)
71
72 def forward(self, input):
---> 73 return nn.functional.interpolate(input, scale_factor=self.scale_factor, mode=self.mode, align_corners=False)

~/anaconda3/lib/python3.7/site-packages/torch/nn/functional.py in interpolate(input, size, scale_factor, mode, align_corners)
2497 if mode in ('nearest', 'area'):
2498 if align_corners is not None:
-> 2499 raise ValueError("align_corners option can only be set with the "
2500 "interpolating modes: linear | bilinear | bicubic | trilinear")
2501 else:

ValueError: align_corners option can only be set with the interpolating modes: linear | bilinear | bicubic | trilinear

Changed layers.basic.Upsample, in forward, the parameter align_corners to None to avoid the error.