NAFNet

NAFNet copied to clipboard

why layernorm so important?

1

我尝试了巨量其他norm或者差不多的方式,比如bn,nfnet的结构,等等,发现只要替代ln网络全挂了. 但是其他的denoise/sr任务并没有这个现象,比如esrgan系列甚至都可以完全不要norm,太神奇了,不知道两年后作者能不能看到这个问题

YilanWang

可以提供单通道的预训练模型嘛，或者告知一下怎么训练单通道模型的方法

Qik-723

I see that you implement yourself the layer normalization instead of using the LayerNorm available in Pytorch https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html Is there a reason to not use the one from Pytorch ?...

edmBernard

在您的摘要中体现的与其他方法计算成本的对比，在论文中并未具体体现。请问您是如何得到这个计算成本对比结果的。感谢！

g483122074

Hi, I've been reviewing the guidelines for setting up the SIDD dataset and have some questions: 1. **Use of Official Dataset for Training and Validation in NAFNet**: - It seems...

hij1112

inference for all the images in one folder is added.

1

the denoising demo was done one by one. With demo_folder script all the images in one folder can be tested. Usage is also simple, --input-folder and --output-folder must be specified...

utkucolak

tengriyizz

作者您好，请问如果想单独把图像复原网络摘出来放到我的目标检测网络里的话，需要看哪个文件下的代码

sunday1112

请问test得到的结果如何恢复到原始图片的大小

1

lwtgithublwt

cuncunsama