Hello, professor. Recentlly I have trained many times with my own T1-T2 weighted MRI dataset on the pix2pix model. First I set the size of pseudo-color image to 512*512, and the train dataset obtains 3166 images with the test dataset obtains 990. Then I crop each image to four part which size is 256*256, so that I can see whether the input size influence the information distribution, however the results seem similar. I have trained almost 100 epoches each time, but the value of G_GAN, D_fake and D_real remains unchanged between [0 2] all the time, and G_L1 stays around 10（lambda=100）after 10th epoch, is this normal?
Finally, I hope you could give me an advice about how to do the validation during the training so that I can avoid overfitting and stop the training. Thank you so much.
the figure are results in 75 epoch，512*512
Hello, professor. Recentlly I have trained many times with my own T1-T2 weighted MRI dataset on the pix2pix model. First I set the size of pseudo-color image to 512*512, and the train dataset obtains 3166 images with the test dataset obtains 990. Then I crop each image to four part which size is 256*256, so that I can see whether the input size influence the information distribution, however the results seem similar. I have trained almost 100 epoches each time, but the value of G_GAN, D_fake and D_real remains unchanged between [0 2] all the time, and G_L1 stays around 10（lambda=100）after 10th epoch, is this normal?
Finally, I hope you could give me an advice about how to do the validation during the training so that I can avoid overfitting and stop the training. Thank you so much.
the figure are results in 75 epoch，512*512