Why use parallel_forward about bicyclegan

[HelenMao]

I also have another question,
half_size = self.opt.batchSize // 2 self.real_A = Variable(self.input_A) self.real_B = Variable(self.input_B) # A1, B1 for encoded; A2, B2 for random self.real_A_encoded = self.real_A[0:half_size] self.real_A_random = self.real_A[half_size:] self.real_B_encoded = self.real_B[0:half_size] self.real_B_random = self.real_B[half_size:]
why the batches are taken apart into two parts, and I find no where to use self.real_A_random as input to the network

[junyanz]

[edited] I think self.parallel_forward is used for MultiGPU. I will refactor it later with DataParallel function.

D_NLayersMulti is an implementation of PatchGAN. The size of final output map depends on the argument n_layers.
I answered your question regarding self.real_A_random in this post.

[HelenMao]

Thanks for your replying

[lc82111]

According to the following net_D architechture, I think D_NLayersMulti implemenates two scale input through AvgPool2d.

  (model_0): Sequential(
    (0): Conv2d(1, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (1): LeakyReLU(negative_slope=0.2, inplace)
    (2): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (3): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (4): LeakyReLU(negative_slope=0.2, inplace)
    (5): Conv2d(128, 256, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (7): LeakyReLU(negative_slope=0.2, inplace)
    (8): Conv2d(256, 512, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
    (9): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (10): LeakyReLU(negative_slope=0.2, inplace)
    (11): Conv2d(512, 1, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
  )
  (down): AvgPool2d(kernel_size=3, stride=2, padding=[1, 1])
  (model_1): Sequential(
    (0): Conv2d(1, 32, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (1): LeakyReLU(negative_slope=0.2, inplace)
    (2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (3): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (4): LeakyReLU(negative_slope=0.2, inplace)
    (5): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
    (6): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (7): LeakyReLU(negative_slope=0.2, inplace)
    (8): Conv2d(128, 256, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
    (9): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
    (10): LeakyReLU(negative_slope=0.2, inplace)
    (11): Conv2d(256, 1, kernel_size=(4, 4), stride=(1, 1), padding=(1, 1))
  )
)