search

Xiaofeng-life / SFSNiD

30 stars 1 forks source link

Yes, I am a bit confused about this. I have noticed that some dehazing tasks use the original size to test PSNR, and the PSNR value of the original size is slightly lower than that of 256 × 256. #8

Closed yushanwendiepisode closed 1 week ago

yushanwendiepisode commented 2 months ago 
          Thank you for your reply. As far as I know, it is traditional operations to train model by resizing the training images, and then use its checkpoint to test images with whole size.  Why are psnr so different when I just change the size of the test images? As you say, I can train from scratch and make the training and test images of the same size, but it's inflexible when training multiple datasets with different input sizes. Besides, for a fair comparison, both the training and the test set in SFSNiD are resized to 256. And are the other comparison methods treated in the same way? I would appreciate it if you could help me out.

Originally posted by @jm-xiong in https://github.com/Xiaofeng-life/SFSNiD/issues/6#issuecomment-2187773651

Xiaofeng-life commented 2 months ago

Yes, I have observed the same problem. The evaluation results may be slightly different for test images of different sizes. We set all algorithms to have the same output size. The reason is that some algorithms cannot accept real-world images of arbitrary sizes as input. In our paper, real-world evaluation is an important part. Perhaps you can resize the output instead of using the experimental results provided in my paper.

yushanwendiepisode commented 2 months ago

Yes, I have observed the same problem. The evaluation results may be slightly different for test images of different sizes. We set all algorithms to have the same output size. The reason is that some algorithms cannot accept real-world images of arbitrary sizes as input. In our paper, real-world evaluation is an important part. Perhaps you can resize the output instead of using the experimental results provided in my paper.

Thank you for your answer. I have another confusion. Did you use 5000 images for each of the NHCL, NHCM, and NHCD datasets? (Excluding 500 bright images from each dataset and using low light images as GT), or only using 500 low light and 500 foggy PNG images, I noticed that other images are in BMP format and some have high contrast and brightness.