zrongcheng

Results 7 comments of zrongcheng

@ChenyangQiQi 对,用来替换掉frequency transform的

@ChenyangQiQi 我只用了fuse的前一帧和当前帧组合的结构,具体的卷积等细节不一样

@xuwenju123 重点了解一下color transofrm和frequency transform,这俩在论文中的结构,其实可以用更简单的结构替换掉;比如avgpool替换的是论文中的LL,不过还需resize进行concat,conv是最后用来降卷积通道的

我知道,公式(9)是符合的,我是说sigma_ll0只有第一帧需要通过a*z_LL+b算,后面帧的sigma_ll0是前面帧传入的

The value of mean is calculated separately for the RGB channels of all pictures. Generally, we ususlly finetune model which trained with the imagenet, so the value of mean was...

Did you solve this bug?

I met the same question, please check this link https://github.com/XPixelGroup/BasicSR/blob/479ec97e8a/basicsr/utils/logger.py with the repo code, just replace it by the link's part