Closed jrmuizel closed 6 years ago
The psychovisual model that PIK uses is not compatible with SSIM, PSNR-HVS-M and many others. One needs to actually look at the images or uses butteraugli to get indicative results.
I'll look into it, last time, I had issue compiling it.
Please let us know what issues you encounter and which compiler - preferably via Github issue :)
@jan-wassenberg Thanks, it is working fine with the latest git.
It seems computationally intensive, I don't know when I will have the time to include it.
@WyohKnott Thanks for letting us know! Yes, we haven't optimized the encoder yet, it will be much faster in future.
shows that for two images pik wins with a huge margin on butteraugli, and gives okeyish results on ssimulacra.
PIK encoding shouldn't be a lot slower than AV1, but there is a bug that leads sometimes to infinite looping at distances smaller than 0.75 or so. Perhaps you were hit by this bug. We have a fix for this and will fix the opensource repo next week.
PIK is not all just about slowness and infinite looping :-) ... PIK's decoding is about 200x faster than AV1, and we have a fast mode that compresses 15 Mpixels/second -- just not accessible by the command line yet. We try to add that too next week to the command line.
I would like to add more context to @jyrkialakuijala comment.
I just refresh the results again with new pik(updated 3 days ago).There was a lot of commit related to performance and the infinite loop occurring at maxerror 0.5 is now fixed.
New result with latest pik : https://encode.ru/threads/2814-Psychovisual-analysis-on-modern-lossy-image-codecs?p=54583&viewfull=1#post54583
As far as encoders are concerned , i used the latest version for each encoder and with the help of the authors i was able to find the optimal flags for each encoder( optimized for best image quality ).
Bash script : https://github.com/khavishbhundoo/Psychovisual-analysis-on-modern-lossy-image-codecs/blob/master/compression_test.sh
Pik has now been added.
Google has a new image codec called pik that they're working on. According to https://encode.ru/threads/2814-Psychovisual-analysis-on-modern-lossy-image-codecs the results are quite good. It would be great to have it included here.