search

chenhsuanlin / spatial-transformer-GAN

ST-GAN: Spatial Transformer Generative Adversarial Networks for Image Compositing :eyeglasses: (CVPR 2018)
MIT License
334 stars 72 forks source link

Is it normal to have loss value around zero? #2

Closed beratkurar closed 6 years ago

beratkurar commented 6 years ago

Hi, We have adapted your code into our images. Could you please guide us whether the following values for the first iterations of the training are normal?

it.670/50000 lr=X(GP),1e-05(D) loss=X(GP),-0.0001(D) norm=1.0000 time=1569.77 it.680/50000 lr=X(GP),1e-05(D) loss=X(GP),-0.0001(D) norm=1.0005 time=1592.73 it.690/50000 lr=X(GP),1e-05(D) loss=X(GP),-0.0001(D) norm=1.0004 time=1615.83 it.700/50000 lr=X(GP),1e-05(D) loss=X(GP),-0.0000(D) norm=0.9995 time=1638.43 it.710/50000 lr=X(GP),1e-05(D) loss=X(GP),0.0001(D) norm=1.0001 time=1661.89 it.720/50000 lr=X(GP),1e-05(D) loss=X(GP),-0.0000(D) norm=1.0000 time=1684.70 it.730/50000 lr=X(GP),1e-05(D) loss=X(GP),0.0001(D) norm=0.9999 time=1708.04

Thank you.

chenhsuanlin commented 6 years ago

Sorry for the late reply. It's hard to tell only from the numbers, but if you're able to run with the default settings there shouldn't be much of a problem :)