Ask for 'Validation' during training and the trend about G_GAN, G_L1, D_fake and D_real

[Paylet]

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

[Gavin666Github]

look， I met the same situation as you.

[happyday521]

Hello, I have the similar situation with you. Have you solved it? I have 10000 images to train my model.
My loss plot is like this:

[Paylet]

I have improved the accuracy of the input, and the LI_loss reduced to 4. But I can't find a good way about the validation in the tain.

[carluri]

Hi @Paylet, were you able to solve your problem? Also, were you training with the default parameters? or did you change them (e.g. learning rate, batch_size, etc)? I'm trying to start something similar and I'm also having a similar issue as described by you. Any help is appreciated. Thank you.