[MPlayer-dev-eng] [LONG] PSNR comparison, new source and results

[D Richard Felker III]

On Fri, Jun 18, 2004 at 02:35:18AM +0300, Ville Saari wrote:
> On Wed, Jun 16, 2004 at 04:15:07PM -0400, D Richard Felker III wrote:
> 
> > BPP is a useless metric because quality does not scale with pixels. If
> > anything it scales closer to sqrt(pixels) so you should measure
> > against bits-per-line or bits-per-diagonal-length.
> 
> Better use bits/sqrt(pixels). Different aspect ratios give very different
> pixel counts for the same line or diagonal length.

BTW a few interesting tests to determine the real correlation:

1. Choose a source clip and encode it at q=2 with various scalings and
   plot sqrt(pixels) against the resulting bitrate.

2. Choose a source clip and encode it with constant-q at various
   scalings, but for each scaling choose the maximal quantizer that
   doesn't look horrible (2 for 320x240, probably 4-5 for 640x480).
   Plot the results the same way.

3. Again choose a source clip. This time try encoding at various
   scalings, and for each scaling, vary the bitrate until you get a
   fixed PSNR value across all scalings. Then plot bits/sqrt(pixels).

Each has its pros and cons. If anyone feels like trying these tests it
would be cool to post results. I'll be away and don't have time
myself.

Rich