zhangyhuaee/KinD

Issues

• evaluate.py generates images different from results of paper

  #13 opened 4 years ago by T5eng
  2

• Fix for 4D color channels: What happens when loading an image with 4 channels instead of 3...

  #16 opened 3 years ago by ChilledFerrum
  0

• There are many halos in the results of the Reflectance Restoration Net training

  #15 opened 3 years ago by ubyang
  0

• How to test images without high data

  #14 opened 4 years ago by eeerpjw
  0

• The NIQE score

  #12 opened 4 years ago by JiaqianLi
  0

• When I test your pre-trained model, I found some problems

  #3 opened 4 years ago by hanxuhfut
  5

• about PSNR and SSIM

  #11 opened 4 years ago by Justones
  1

• The size of the input image

  #10 opened 5 years ago by zhangxiaopang88
  0

• I have some questions when I run evaluate.py

  #9 opened 5 years ago by Vampire0716
  0

• I have some questions when I run evaluate.py?

  #8 opened 5 years ago by Vampire0716
  0

• Please share latest model in google drive!

  #4 opened 5 years ago by ANSHUMAN87
  1

• I wonder how to find NPE, MEF and LIME dataset?

  #7 opened 5 years ago by anothersin
  0

• 你好，感谢分享你的代码。我发现下载的训练数据中低光照图像中并没有对应的正常光照图像，比如79.png。因此按照你程序里的直接读取两个目录下的图像列表并排序得到的图像对并不能保证低光照和强光照图是同一张对应的。我修改数据集中的问题后训练效果反而更差了，请问这是怎么回事？

  #5 opened 5 years ago by changruowang
  0

• Here are several mistakes on the correspondance in "our485", like the 79.png, 735.png. The order of low and high images in "our485" do not match, like the 79.png and 111.png. But I correct it ,the decom net train result seem to bee worse! How do you solve it ?

  #6 opened 5 years ago by changruowang
  1

• The result of training code is different form the checkpoint you provided

  #2 opened 5 years ago by lvxiaoqian
  1