Size mismatch RuntimeError on training with FINE_SIZE more than 256.

[siddhartha-iitd]

Using LOAD_SIZE=640, FINE_SIZE=256 for training works fine. However, using LOAD_SIZE=640, FINE_SIZE=512 or LOAD_SIZE=512, FINE_SIZE=512 throws a RunTimeError shown below:

Traceback (most recent call last):
  File "./train.py", line 34, in <module>
    model.optimize_parameters()
  File "bicyclegan/BicycleGAN/models/bicycle_gan_model.py", line 209, in optimize_parameters
    self.forward()
  File "bicyclegan/BicycleGAN/models/bicycle_gan_model.py", line 106, in forward
    self.z_encoded, self.mu, self.logvar = self.encode(self.real_B_encoded)
  File "bicyclegan/BicycleGAN/models/bicycle_gan_model.py", line 83, in encode
    mu, logvar = self.netE.forward(input_image)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/parallel/data_parallel.py", line 121, in forward
    return self.module(*inputs[0], **kwargs[0])
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "bicyclegan/BicycleGAN/models/networks.py", line 602, in forward
    output = self.fc(conv_flat)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/container.py", line 91, in forward
    input = module(input)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/linear.py", line 55, in forward
    return F.linear(input, self.weight, self.bias)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/functional.py", line 1024, in linear
    return torch.addmm(bias, input, weight.t())
RuntimeError: size mismatch, m1: [1 x 1024], m2: [256 x 8] at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:249

Could you please help me find out how can I train the model for FINE_SIZE values of 512 or more ? Thanks!

[junyanz]

I think the question was posted in the wrong repo. This question is related to BicycleGAN. The default generator in BicycleGAN is unet_128 or unet_256. It now only supports 128 or 256. For 512 images, you can either use resnet_256 or add a few layers to unet_256.

[siddhartha-iitd]

Thanks @junyanz for reply. I just made that change and it works :) Thanks again!