Difference between psnr calculation in torchmetrics and piq

[bojobo]

Hi!

Recently I've switched from using the piq for my metrics to torchmetrics.
After this switch, I've noticed that the reported psnr somehow turned negative.
After investigating I've found the reason:
Here it can be seen that piq computes the psnr without the first part of the subtraction (compare with the torchmetrics implementation.

What is the reason for this difference?

[bojobo]

I've also asked a similar question over at photosynthesis-team/piq.

[bojobo]

Nevermind. They first scale the tensors to [0, 1] making the first part of the subtraction 0.