JM 10.1 vs x264 rev 382 encoder: PSNR comparison

example JM config: encoder.cfg
summary of JM opts: high profile, fast motion estimation, full RD, 1 B-frame, 5 references

example x264 commandline:
x264 --qp 24 --bframes 1 --ref 5 --8x8dct --subme 6 --trellis 2 --me umh --b-rdo --mixed-refs --direct spatial --no-b-adapt --keyint 999 --scenecut -1 foreman_352x288.yuv
(i.e. same bitstream features as JM used)

For the purpose of comparison, I disabled ratecontrol in both codecs (leaving constant QP), and disabled adaptive frame-type decision in x264.
The QPs tested are: 16, 20, 24, 28, 32, 36

All clips are approximately 250 frames, CIF resolution. In all plots, the better lines are the above options, and the worse lines are the same except with no B-frames and 1 reference.

Total encode time on an Athlon-XP 2500, for 24 videos * 6 bitrates * ~250 frames:
with B-frames & refswithout
JM 202437 sec153032 sec
5.817 sec/frame4.397 sec/frame
x264 w/o trellis
quantization
4889 sec3047 sec
0.1405 sec/frame0.0876 sec/frame
x264 7748 sec6256 sec
0.2226 sec/frame0.1798 sec/frame

Average of all of the below
Absolute rate vs distortion (higher is better) Normalized (lower is better)


akiyo

bridge-close

bus

coastguard

container

flower

football

foreman

hall

highway

mobile

mother_daughter

news

paris

silent

stefan

tempete

tennis

waterfall

Etorofu

Exile

Haibane

Hpii

Shell


Last modified: 2005-12-5