It took me a while to understand Lpips in usage because strange and ugly stuff seems to happen always.
I went on with experimenting because my normal ms-ssim seemed to be not very effective in training with clipV
So i did Logcosch + Lpips vgg16 and i was amazed.
The first 3000-5000 its you only see very ugly faces and you wonder if it will every succeed.
at 10000 its you see sharp but somewhat strange faces.
and then it very quickly gets very good. You just have to ignore the beginning strangeness.
For now this experience is only with clip, but with this loss function combo the results are way better with a very reduced training time then using MS-SSIM loss function.
When combining MS-SSIM with Lpips vgg16 also gives much slower training results.
Vgg is way more effective then squeeze
The level of detail i get with clipv on this loss function way exceeds what i ever got with other loss functions.