search

Ha0Tang / SelectionGAN

[CVPR 2019 Oral] Multi-Channel Attention Selection GAN with Cascaded Semantic Guidance for Cross-View Image Translation
http://disi.unitn.it/~hao.tang/project/SelectionGAN.html
460 stars 63 forks source link

Checkers on generated images #16

Open zhangdan8962 opened 3 years ago

zhangdan8962 commented 3 years ago

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! Screenshot from 2021-09-06 01-18-38

Ha0Tang commented 3 years ago

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

zhangdan8962 commented 3 years ago

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

Ha0Tang commented 3 years ago

Yes, you are right.

zhangdan8962 commented 3 years ago

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 commented 2 years ago

It should be for a fair comparison with existing methods.

zhangdan8962 commented 2 years ago

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