I’ve a question about Multiscale SSIM implementation. In the code (
torchmetrics/src/torchmetrics/functional/image/ssim.py at v1.4.0.post0 · Lightning-AI/torchmetrics · GitHub),
I see that we are ignoring the last contrast_sensitivity. I looked up the Multiscale SSIM paper (https://www.cns.nyu.edu/pub/eero/wang03b.pdf) and I understand that we need the luminance from the last hierarchy level. But I don’t see that it says to replace the last contrast_sensitivity with luminance. So, I want to know why that is done. thanks !
I figured it out. contrast_sensitivity is getting replaced with ssim, which is alright.