View Single Post
Old 10th July 2016, 06:04   #1035  |  Link
youli
Registered User
 
youli's Avatar
 
Join Date: Mar 2015
Location: Ukraine
Posts: 23
Quote:
Originally Posted by r0lZ View Post
You may also use the SSIM or PSNR tunes to have a relatively precise and objective measure of the quality of the encodings. You should encode the same short clip several times with different settings, and compare the final SSIM or PSNR values to determine the "best" method...
Yes, you are right. But...

If the "psy" and "aq" optimizations are enabled, then they will reduce the level of SSIM, and also misrepresent the level of PSNR. Rather, these values will be assessed incorrectly.
I mean this options : "--rd, --psy-rd, --rdoq-level, --psy-rdoq" and "--aq-mode, aq-strength".

Correct results of PSNR and SSIM can only be obtained if psycho-visual optimizations and adaptive quantization are disabled.

Of course I assessed structure similarity during the tests. If crf = 23, the SSIM Mean about 98.9%.

For x264 "Avg QP" for B frames recommended less than 23.
It can be assumed, that for x265 "Avg QP" for B frames recommended less than 28.

x265 [info]: frame I: 560, Avg QP:20.08 kb/s: 29462.21
x265 [info]: frame P: 81439, Avg QP:23.06 kb/s: 10158.44
x265 [info]: frame B: 55969, Avg QP:25.15 kb/s: 2148.81

So above, I have a slight excess of the bitrate at --crf 23.
youli is offline