What happens when changing SeD to SeG?
Opened this issue · 3 comments
I have an interest in the application of semantic information. What happens when semantic information is added to the generator? Have you done any experiments on this?
Hi, thanks for your interests! We discuss this problem in the supplementary. We changing SeD to SeG, the performance drops because semantics can no longer obtained from a high-resolution input. (As the generator takes low-resolution image as the input)
You should to see "SFTGAN"
Hi, thanks for your interests! We discuss this problem in the supplementary. We changing SeD to SeG, the performance drops because semantics can no longer obtained from a high-resolution input. (As the generator takes low-resolution image as the input)
HI, the xintao's work "SFT-GAN" in 2018 has used the semantics to generate more accurate texture.