PatchGAN?

[haoliangjiang]

I think the original pix2pix GAN is using patchGAN. Where is the code related to it in your Tensorflow version?

[32nguyen]

I have the same question. haoliangjiang. May I ask you a question. I found something that is not quite clear. Let take example on facades data: with the function: def load_image(image_path): input_img = imread(image_path) w = int(input_img.shape[1]) w2 = int(w/2) img_A = input_img[:, 0:w2] img_B = input_img[:, w2:w] return img_A, img_B

Then they are concatenated in order AB by: img_AB = np.concatenate((img_A, img_B), axis=2)

so the model is about generating the realistic image from A to B (left to right in the image.). However, in "build_model(self)", A and B is flipped to each other by:

self.real_B = self.real_data[:, :, :, :self.input_c_dim] self.real_A = self.real_data[:, :, :, self.input_c_dim:self.input_c_dim + self.output_c_dim]

My understanding is real_B is now the input imageA. So the calculation of "tf.reduce_mean(tf.abs(self.real_B - self.fake_B))" does not make any sense.

Bests,

[praveenkumarchandaliya]

In the case of PatchGAN, last layer of Discriminator have output channel always 1. nn.Conv2D(inchannel,outchannel,kernelsize,stride,padding): nn.Conv2D(128,1,3,1,1) ' '

[Xuecheng-anhui]

I think the original pix2pix GAN is using patchGAN. Where is the code related to it in your Tensorflow version?

In his code, is not a picture outputting a value after D, but number of channel values,.which corresponds to patchGAN.

[maryam089]

Did anyone find where is PatchGAN in this code? the last layer of the discriminator has feature maps 16x16x512 .... Is this size of feature maps refers to PatchGAN idea ? as mentioned in here https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/issues/39#issuecomment-305575964