About the epochs in Training

AAnoosheh / ToDayGAN

http://arxiv.org/abs/1809.09767

BSD 2-Clause "Simplified" License

172 stars 32 forks source link

About the epochs in Training #20

Closed seainlost81 closed 4 years ago

seainlost81 commented 4 years ago

Anoosheh, thank you for your work and provided code. I have a question about the epochs in the training. As stated in your article，training is run for 40 epochs. So, when we try to train a model as yours, the 'num_epochs_constant_LR' in 'python train.py --name --dataroot ./datasets/ --n_domains --niter --niter_decay ' is 20 or 40? Because when we use 20, the effect is not very satisfactory. Looking forward to your reply. Thanks in advance. 1418235223450115_fake_0 1418235223450115_real_1

AAnoosheh commented 4 years ago

Wow I totally forgot to update the number of epochs in the paper. I think I used to do 40 at first, and then for my final model I actually ran it for 150 (75+75) and then I didn't update the text.

seainlost81 commented 4 years ago

Anoosheh, thank you for your prompt reply.