Clamp image before PSNR and MS-SSIM

[Freed-Wu]

🚀 Feature

from torchmetrics.image.psnr import PeakSignalNoiseRatio

metric = PeakSignalNoiseRatio(clamp=[0, 1])
for inputs, targets in dataloader:
    outputs = model(inputs)
    metric(outputs, targets)

Then we will clamp images to $[0, 1]$, and calculate data_range = 1 - 0.

Motivation

Image have a legal range, $[0, 255]$ or $[0, 1]$, so a clamp is necessary for calculate PSNR.

Pitch

Alternatives

from torchmetrics.image.psnr import PeakSignalNoiseRatio

metric = PeakSignalNoiseRatio(data_range=[0, 1])
for inputs, targets in dataloader:
    outputs = model(inputs)
    metric(outputs, targets)

if data_range is a int, it keep same behaviour as before. if it is a list, will clamp image to data_range and calculate its delta as data_range.

Additional context