search

Zhangjinso / PISE

123 stars 28 forks source link

About the LPIPS #4

Closed happyday521 closed 3 years ago

happyday521 commented 3 years ago

Hi， great work！ I have a problem about the results of LPIPS. Since you split the dataset with the same way as GFLA , why are the LPIPS results in your paper(GFLA: 0.2219, PATN:0.2520) different from that reported in the paper of GFLA?(GFLA: 0.2341, PATN:0.2533). What is your test setup？ Thanks！

Zhangjinso commented 3 years ago

Hi, we compute lpips using https://github.com/richzhang/PerceptualSimilarity, and the inputs are cropped to 256X176 due to the valid background. For the results of GFLA, most methods as well as ours use input images of size 256X176 and padding to 256X256. The input in the paper of GFLA is from high-resolution images and resized them directly to 256X256, which contain more information compared with low-resolution images. In our paper, we take 256X256 images resized from 256X176 as input of GFLA, which can be the main reason.

happyday521 commented 3 years ago

Got it. Thanks for your reply.