Open DequanZhu opened 4 years ago
你好,我最近也在跑这个代码。可以加你交流一下吗?我的微信:loveanshen 我的QQ:519838354 我的邮箱:519838354@qq.com 非常期待你百忙中的回复
probably because it avoids usage of edges in the masked part by keeping it blank, since this model uses edge estimation and then inpainting it helps the output, other models such as pix2pixhd use just the inpainting part .
I agree with your opinion@DequanZhu,The mask can not always be knew, if mask must be provided, I think this methods lacks generalization.
It's hard to diagree with your viewpoint! @DequanZhu
Why must provided a mask if I want to repair a damage image, I mean sometimes I don't know which part is damaged, or damage part is everywhere. I have read some papers about inpainting, most networks in these papers must provided mask for training but no need for test mode. These networks can determine which part should to be repaired which means can repair damaged image automatically and yours must provided masks for every image.