XingangPan/deep-generative-prior

Issues about switch off the update G

wtybest opened this issue · 3 comments

wtybest commented

When I switch off the options --update_G, to perform the reconstruction, which means only optimize the latent code, the result seems far from satisfactory, even for imagenet val datasets.
image
In my understanding, optimizing just for latent code might not be perfect, but at least it shouldn't be this bad.
Is that consistent with what you tried?

XingangPan commented

@wty-ustc Yes, the result is consistent with what I tried.
Optimizing just for latent code would produce very inaccurate reconstruction for BigGAN, even if you enlarge the learning rate. That's why we need to optimize the generator too.

wty-ustc commented
zhangxiangyu19 commented

it's the same with my test. How do other generators do it, is it because the training set is smaller?