Mask-ShadowGAN icon indicating copy to clipboard operation
Mask-ShadowGAN copied to clipboard

Published 20 hours ago • verified publisher icon xw-hu

Unable to convergence

Open handsomefengfeng opened this issue 4 years ago • 4 comments

I've trained the model on the given USR dataset.

None parameters were changed except the dataroot.

However, the generator was unable to convergence.

image

Meanwhile, the discriminator convergence better than the generator. image

handsomefengfeng avatar Apr 11 '20 03:04 handsomefengfeng

Could you please provide any solution or suggestion? Thanks!!!

handsomefengfeng avatar Apr 11 '20 03:04 handsomefengfeng

Could you please provide any solution or suggestion? Thanks!!!

I have the same problem, have you solved it yet?

1214710638 avatar Nov 09 '20 08:11 1214710638

How about re-starting training? Maybe the initial random seed is not good.

xw-hu avatar Nov 09 '20 08:11 xw-hu

How about re-starting training? Maybe the initial random seed is not good.

The above is the result I re-trained on the provided USR dataset without any configs changed in the train_MaskShadowGAN.py.

You mean it's the unsatisfactory of initial random seed that make the loss not so smooth. Could you please tell how to make the initial random seed appropriate?

Thanks~~! :)

handsomefengfeng avatar Nov 10 '20 13:11 handsomefengfeng