about the Training Strategy

[riaris]

I used the code to train a new model without changing hyperparameter about batch_size lambda etc, when it finished at 600 epoch, the generated images onlg got 4.20±0.15 for inception score not the 4.36±.03 at paper. the problem was caused by Training Strategy what I didn‘t know or the init of netG and netD?

[qizhongjian]

Me too. @riaris @taoxugit