[Ffmpeg-devel] H.264 decoding bug in either lavc/JM (x264 too?)

Tuukka Toivonen tuukkat
Mon Jul 25 19:33:38 CEST 2005

On Mon, 25 Jul 2005, Loren Merritt wrote:

>> x264 command line:
>> --bframes 2 --8x8dct --b-pyramid --qp 26 --direct spatial
>> The problem occurs only if using all three options --8x8dct, --b-pyramid,
>> and --direct spatial. Most other options seem not to matter.
> I can't see what 8x8dct would have to do with it.

You're correct, it seems that even without --8x8dct
my test video on the web has the problem. I only tried removing
--8x8dct once, and the problem appeared to go away then. Maybe
I was just (un)lucky or something.

If changing --bframes 2 to just 1, the problem
goes away with the test video (just verified).

>> For now, for encoding with x264, --direct temporal
>> should be used (it increases bit rate around 3% at fixed
>> PSNR, but removing --8x8dct or --b-pyramid has larger
>> impact, 4-5%).
> btw, I think the 3% penatly goes away if you use --ref > 1.

No, I used actually 6 reference frames with a command line like this:

/usr/local/bin/mencoder -quiet -endpos 10 -rawvideo 
on:format=i420:fps=25:w=640:h=480 -ovc x264 -x264encopts 
:me_range=32:subq=5:nochroma_me:chroma_qp_offset=2 -o test.25860.avi 

Averaged over my 12 videos, I got these results:
method x264dp2-245 any    bitrate, psnr 35.68 .. 39.24 relative bitrate: 103.04773
method x264dp2-245 low    bitrate, psnr 35.68 .. 36.87 relative bitrate: 103.29744
method x264dp2-245 high   bitrate, psnr 38.06 .. 39.24 relative bitrate: 102.83746

(compared to x264dp1-245 where dpX means direct_pred=X).

Here's a graph: :)

More information about the ffmpeg-devel mailing list