Open Ying156209 opened 5 years ago
Hi,
Usually, the checkerboard artifacts will be mitigated after longer training. Or you could try to replace deconv operation by other operations like upsampling (as stated in https://distill.pub/2016/deconv-checkerboard/)
Hi,
Thanks for the brilliant work and open source code. I ran the code with my own dataset, and I find that artifacts occurred sometimes. I also ran your trained model on summer-winter dataset and found the grid-like artifact still. Is there any idea why this happened?