Training loss does not decrease

[nguyenpbui]

Dear authors @junyanz and @taesungp,

First, thank you for your excellent work!

I am currently using the pix2pix model to predict future images (gray-scale) based on the current one in the medical field. I modified the code to use 7 input channels and 1 output channel for Generator.

During the training, I observed that the loss values of G_GAN, G_L1, D_Real, and D_Fake have not changed much since the beginning of the training process. (Figure 1)

Regarding the training results, the fake_B images are very different from the real_B ones.

I am new to this task and any insights/explanations/recommendations are highly appreciated! Thank you very much.

[setarekhosravi]

Hello @phuocnguyen2008 I have the same problem, did you get the answer?

[tongfh1]

Hello, how do you modify the code to realize the 7 input channels and 1 output channel?

[Parth527]

@tongfh1 there are flags input_nc and output_nc for that purpose