Artoriuz
commented
3 years ago To be honest with you I didn't intend to downscale in linear light anywhere (and I think this is the only place where I fucked up, the subsequent downscales for the other sections were all done in imagemagick). I left it like that because I didn't think it would change the order of the scalers, but I'll try to fix it today.
About NIQE, it's just for convenience. Matlab comes with it built in so adding it to the mix was trivial. I'm trying to keep things simple this time (so I'm able to update it more quickly) but if you have some suggestions of good no-reference metrics with matlab implementations I'd be happy to include them.
On Sun, Dec 6, 2020, 7:40 AM igv notifications@github.com wrote:
Compare SSSR or any scaler like spline36 before and after the violet upscaling remake. They are more aliased now.
Also why are you using such an old NR metrics? For example there is a newer version of NIQE called IL-NIQE http://www4.comp.polyu.edu.hk/~cslzhang/IQA/ILNIQE/ILNIQE.htm. Or BMPRI https://drive.google.com/file/d/1C_NxTLvnBOJDGhqqtixCkra0LIMP6loF/view. There is also a very good deep learning based FR metric http://civc.ucsb.edu/graphics/Papers/CVPR2018_PieAPP/ (windows executable available).
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Artoriuz/artoriuz.github.io/issues/6, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC4EY4MUY7Q3MJCSBXVQITLSTNNRPANCNFSM4UPHF5NQ .
Compare SSSR or any scaler like spline36 before and after the violet upscaling remake. They are more aliased now.
Also why are you using such an old NR metrics? For example there is a newer version of NIQE called IL-NIQE. Or BMPRI. There is also a very good deep learning based FR metric (windows executable available).