Checkers on generated images

[zhangdan8962]

Hi,

I have tried to duplicate the result based on your dayton_a2g_256_pretrained model. However, there are many checkers in the generated images which is not clearly shown in Fig.4 of you paper. I followed all the hyperparameters as you mentioned in readme file

I am wondering if you encountered the same issue in your experiments or do I miss anything?

The attached is one of the screenshots. Thank you in advance!

[Ha0Tang]

The last image "I" is the final result. Are you talking about this result？

[zhangdan8962]

My bad. I thought fake_B is the final result, which should be the result from stage1 right?

[Ha0Tang]

Yes, you are right.

[zhangdan8962]

And one last question regarding the batch size. Is it simply because the cuda memory limit or is there any reasons behind the fact that you were using batch size 4?

[Ha0Tang]

It should be for a fair comparison with existing methods.

[zhangdan8962]

I also realize that the fake_D is pretty blurry. Is it the output of stage 1 or 2?