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

[Romain Dolbeau]

Jan Knutar wrote:

> I've never so far when I've encoded myself and done some limited PSNR 
> tests first, seen that qpel would improve things. Would it be possible 
> that you include qpel or not in further tests? I'm not convinced of its 
> usefulness yet :D

You're right. On the Matrix scene I get very different result.
I kept vme=4, mbd=2, *cmp=2, mv0, v4mv, and a bitrate od 650 kbps
I then tried *dia=-2 or 1, qpel or not, trell or not, and
0 or 1 b-frames. Basic stuff.

[vmax_b_frames=1 vs. vmax_b_frames=0] Average: from 44.9413 to 44.9138 (-0.0275, -0.0612%) [16 samples] bestDiff: 0, worstDiff: -0.06 pCount: 0, nCount: 12 AvgTimeCost: 65.1324 seconds or 40.2591 %
[qpel vs. null] Average: from 44.965 to 44.89 (-0.075, -0.1668%) [16 samples] bestDiff: -0.05, worstDiff: -0.09 pCount: 0, nCount: 16 AvgTimeCost: 93.6701 seconds or 63.499 %
[trell vs. null] Average: from 44.82 to 45.035 (0.215, 0.4797%) [16 samples] bestDiff: 0.25, worstDiff: 0.18 pCount: 16, nCount: 0 AvgTimeCost: 43.8086 seconds or 25.4044 %
[[dia=1, predia=1] vs. [predia=-2, dia=-2]] Average: from 44.9388 to 44.9163 (-0.0225, -0.0501%) [16 samples] bestDiff: -0.01, worstDiff: -0.04 pCount: 0, nCount: 16 AvgTimeCost: -17.1286 seconds or -8.4413 %

qpel not only raise the execution time by over 63% on
average, it also decrease the PSNR in all case. trell
isn't anywhere as expensive, and it does give you a
nice improvement all the time. *dia=-2 is a tiny bit
better and not much more expensive tha *dia=1.

Matthias Wieser wrote:

 > You did record the time each combination of options has taken, didn't you?
 > I think the "best of the best" is not as interesting as "nearly the best
 > of the best" options with reasonable FPS.

I do now :-) In those 16 cases, the best was 33% slower
than the third best, for a very small gain in chroma
only (order are Y / Cb / Cr / All / user time):

vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=1/mbd=2/qpel/mv0/v4mv/vbitrate=650           43.72   47.33   49.01   44.75   4m17.074s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=1/mbd=2/qpel/mv0/v4mv/vbitrate=650         43.76   47.35   49.02   44.79   4m40.334s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=0/mbd=2/qpel/mv0/v4mv/vbitrate=650           43.78   47.33   48.98   44.80   2m43.024s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=1/mbd=2/mv0/v4mv/vbitrate=650                43.79   47.33   48.99   44.81   2m12.920s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=0/mbd=2/qpel/mv0/v4mv/vbitrate=650         43.80   47.34   48.99   44.82   2m54.140s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=1/mbd=2/mv0/v4mv/vbitrate=650              43.83   47.35   49.00   44.84   2m33.123s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=0/mbd=2/mv0/v4mv/vbitrate=650                43.86   47.32   48.95   44.87   1m43.337s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=0/mbd=2/mv0/v4mv/vbitrate=650              43.88   47.32   48.96   44.88   1m55.608s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=1/mbd=2/qpel/mv0/v4mv/trell/vbitrate=650     43.96   47.43   49.06   44.97   4m58.093s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=0/mbd=2/qpel/mv0/v4mv/trell/vbitrate=650     43.99   47.44   49.05   44.99   3m30.329s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=0/mbd=2/qpel/mv0/v4mv/trell/vbitrate=650   44.00   47.44   49.05   45.00   3m42.789s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=1/mbd=2/qpel/mv0/v4mv/trell/vbitrate=650   44.00   47.46   49.08   45.00   5m23.692s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=1/mbd=2/mv0/v4mv/trell/vbitrate=650          44.07   47.44   49.05   45.06   2m54.441s
vme=4/cmp=2/precmp=2/subcmp=2/dia=1/predia=1/vmax_b_frames=0/mbd=2/mv0/v4mv/trell/vbitrate=650          44.10   47.40   49.00   45.08   2m27.062s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=0/mbd=2/mv0/v4mv/trell/vbitrate=650        44.12   47.40   49.00   45.09   2m37.976s
vme=4/cmp=2/precmp=2/subcmp=2/dia=-2/predia=-2/vmax_b_frames=1/mbd=2/mv0/v4mv/trell/vbitrate=650        44.10   47.46   49.06   45.09   3m15.647s



-- 
Romain Dolbeau