You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi,
I have the exact same issue.
I have installed the conda environment as instructed and trained the network according with the default parameters and it doesn't converge to anything logical at any point.
Hi, all. The master branch might collapse when resolution is larger than 128x128. If you want to obtain similar result in README, you can fall back to this commit, however, you should notice that some ops were not correctly implemented. Besides, you'd better use a lower learning rate, 1e-4 would be fine.
I trained the network with following setting:
python train.py --gpu 0 --train_kimg 600 --transition_kimg 600 --beta1 0 --beta2 0.99 --gan lsgan --first_resol 4 --target_resol 256 --no_tanh --exp_dir /exp
I am using the new CelebaHQ dataset from nvidia.
and I got following stabilization image after 75000 steps:
![128x128-stabilize-074999](https://user-images.githubusercontent.com/11586277/35627825-72ae32cc-069a-11e8-9636-36803fe20fd0.png)
The results on all samples (at various sizes) are similar. I am using python2.7
Anyone experiencing similar samples?
Edit: Addition
I tried python3, using train_no_tanh. Still, the image is same. There is no change.
The text was updated successfully, but these errors were encountered: