search

xiaofeng94 / RefineDNet-for-dehazing

The Pytorch implementation for "RefineDNet: A Weakly Supervised Refinement Framework for Single Image Dehazing" (TIP 2021)
65 stars 16 forks source link

evaluate metrics #10

Open guigui6666 opened 1 year ago

guigui6666 commented 1 year ago

Hi, Can I have your .py file to evaluate metrics such as psnr, ssim, VSI, VI, and SI I don't know how to evaluate the VSI, VI and SI values and my results of ssim is not as good as yours

xiaofeng94 commented 1 year ago

Hey, we used evaluation code from this repo. Please also check your environment and make sure correct dehazed images are generated.

guigui6666 commented 1 year ago

Thank you for your reply. 我注意到这套代码中ssim的测量也是通过调用skimage.measure 的compare_ssim得到的，但不知道为什么我的测量值只有0.69左右，而文章中有0.9+，代码部分我没有改动，我不知道哪里有问题，或许你可以给我提供一些思路么？

xiaofeng94 commented 1 year ago

Hey, 可以尝试以下办法，

  1. 按照this repo测试
  2. 可视化去雾结果，确保有模型有正确的输出。0.69的ssim意味着得到的去雾结果基本无效
  3. 检查skimage.measure的compare_ssim接口文档，确保正确的使用