I got better result than what is reported in the paper.

[kwea123]

Hi, I adopt your MFN (GaborNet) in my coordinate-mlp repo, and conducted the generalization experiment on Natural dataset, and found that it produces a lot better than what is reported in the paper. More precisely, I got 28.75 mean PSNR whereas you got only 26.18. Do you have the detailed training parameters or the training logs (preferably per-scene)? I have mine and am happy to share with you (I will upload to my repo), I wonder what makes the difference.

My worst generalization is for this image (right is network prediction):

But even for this it has 23.24 PSNR.

[kwea123]

I got the wrong evaluation process, sorry. The test was conducted on the other 1/4 pixels, not the whole image...

After correcting this, my score is 26.23 which is close to your result.