question about msssim loss: output = torch.prod(pow1[:-1] * pow2[-1])

[GarveyHung]

Hi. It's a great work.

But, it may be a wrong implement of MS-SSIM.

pow1 = (mcs1, mcs2, mcs3, mcs4, mcs5)
pow2 = (mssim1, mssim2, mssim3， mssim4， mssim5)

let along weights.

The MS-SSIM should be:

MS-SSIM = mcs1mcs2mcs3mcs4mssim5

The implement in your code torch.prod(pow1[:-1] * pow2[-1]) would let to:

MS-SSIM = mcs1mcs2mcs3mcs4(mssim5 ** 4)

Please check if the mistake is mine. Thanks.