[x264-devel] x264 SSIM analysis

[BugMaster]

On Mon, 24 Feb 2014 19:00:00 +0100, Niccolò Belli wrote:
> Hi, I did analyse SSIM from two videos compressed with x264 using 
> CRF=18 and different presets. I expected to get a similar SSIM 
> considering x264 tunes for SSIM by default and not for PSNR. 
> Unfortunately SSIM differs quite a lot (see attached image).

> I did calculate the SSIM using qpsnr (http://qpsnr.youlink.org/):
> ./qpsnr -a avg_ssim -o blocksize=16:fpa=24 -r ~/placebo-lossless.mkv 
> ~/placebo-cfr18.mkv ~/veryfast-cfr18.mkv > output.csv

> Why did I get such a strange result?

> Niccolò

Because nowhere and never there was any official statemantes that same
CRF will produce same PSNR/SSIM or same subjective quality with
different settings (presets). If you thought so than you was wrong or
was misleaded. All what CRF mode mean is constant ratefactor (notice
it is ratefactor and not quality which are not equal thing). And
constant ratefactor means more or less constant tradeoff bitween
quality/bitrate along the the source (not between different sources or
different options).