NAFNet

NAFNet copied to clipboard

Unable to generate the log file.

4

When trying to run the train.py file, I'm unable to generate the .log file. When I tried to debug looks like logger.hasHandlers is always "true", but when I tried to...

Mohan2351999

Hi, I tried running the model locally (using CPU only) but my memory ran out. So with the same source image I tried running it on databricks. I have stored...

Meenakshi-10

Performance of deblurring images which has high gaussian and median blur spots

1

Hi team, Great and thanks for the work team! My dataset involves a mix up of gaussian blur, fog, median blur and few cases of motion blur, Infact in some...

vandana1302238

The possibility of training NAFNet on gaussian-noise images

2

Hi, I was wondering if NAFNet could be trained on gaussian-noise images. If so, is there any pipeline I could follow? Thanks in advance.

shengyenlin

Dear authors, I'm wondering if why you implemented mean shift but not used? Thanks!!

unacanal

复现精度

2

请问作者大大，我浏览了以前的问题看到：要复现文中的精度的话witd32中的bacth size=32 ,iters=20K,lr= 1e-3不能变。请问是bacth size和iters必须一模一样，还是二者乘积数一样就可以了。

wangxiaoxiao6

您好！我注意到在NAFSSR的论文里用到的是L1损失，但是代码里却是用的MSEloss，想知道到底用的哪个呢？期待您的回复

LinPeiMing

请问enc_blk_nums: [1, 1, 1, 28]可以理解为有四个编码器模块，第一二三个编码器模块包含1个NAFBlock，第四个编码器模块包含28个NAFBlock吗？请问这样设计的原因是什么呢？

Amy-214

Process whole folder but no one by one image?

1

Could someone please make a notebook file which denoise whole folder and not one by one image? Thanks

zelenooki87

training nafnet_sidd_width_32 without BasicSR

2

hello, I have trained the nanfet-width32 on SIDD, the option file was the same as yours. The different parts are as follows. - GPU: 8 x TitanXP - I use...

lianghao2000