[Ffmpeg-cvslog] CVS: ffmpeg/libavcodec mpegvideo.c,1.496,1.497

Corey Hickey bugfood-ml
Thu Dec 29 03:30:07 CET 2005


Michael Niedermayer wrote:
> Hi
> 
> On Tue, Dec 27, 2005 at 09:19:03AM -0800, Corey Hickey wrote:
> 
>>Michael Niedermayer CVS wrote:
>>
>>>Update of /cvsroot/ffmpeg/ffmpeg/libavcodec
>>>In directory mail:/var2/tmp/cvs-serv20080/libavcodec
>>>
>>>Modified Files:
>>>	mpegvideo.c 
>>>Log Message:
>>>improved lambda/qp guessing for motion estimation RD in first pass
>>>this only affects encoding with bframes and rate control
>>
>>-----------------------------------------------------------------
>>for i in 1:turbo 2 ; do
>>  time mencoder ~/dumpstream/matrix.vob -aid 128 -oac copy \
>>  -vf crop=718:356:0:60,scale=640:272 -sws 9 -ovc lavc -lavcopts \
>>vcodec=mpeg4:vbitrate=581:psnr:vpass=$i:mbd=2:mv0:trell:cbp:\
>>precmp=2:cmp=2:subcmp=2:predia=2:dia=2:preme=2:vme=5:v4mv:\
>>last_pred=2:vqcomp=0.6:vmax_b_frames=2:qpel -ofps 24000/1001 \
>>  -o test.avi
>>done
>>
>>-----------------------------------------------------------------
>>before:
>>
>>(pass 1)
>>PSNR: Y:40.03, Cb:44.35, Cr:44.66, All:41.08
>>user    34m38.232s
>>
>>(pass 2)
>>PSNR: Y:42.11, Cb:45.17, Cr:45.86, All:42.98
>>user    188m52.138s
>>
>>-----------------------------------------------------------------
>>after:
>>
>>(pass 1)
>>PSNR: Y:40.05, Cb:44.34, Cr:44.64, All:41.09
>>user    34m55.052s
>>
>>(pass 2)
>>PSNR: Y:42.10, Cb:45.17, Cr:45.85, All:42.97
>>user    190m5.245s
>>
>>-----------------------------------------------------------------
>>
>>As is to be expected, low motion scenes look identical. I thought high
>>motion scenes looked nearly identical as well, until I ran into a couple
>>places that looked much worse after this patch. I'm uploading four
>>representative screenshots (two before, two after) to:
>>
>>http://fatooh.org/files/sstemp/
>>
>>I'm on a slow dial-up connection right now, so this message might reach
>>the mailing list a few minutes before the upload finishes.
>>
>>I haven't checked any later commits; these problems might disappear when
>>I try them.
> 
> 
> yes, i hope so too, but this needs to be verified (=test latest cvs
> vs. latest cvs with this change reversed)

I finally caught up to the present and tried this today.

-----------------------------------------------------------------
current cvs:

(pass 1)
PSNR: Y:40.05, Cb:44.29, Cr:44.59, All:41.08
user    34m52.509s

(pass 2)
PSNR: Y:42.04, Cb:45.15, Cr:45.84, All:42.91
user    190m8.100s

-----------------------------------------------------------------
current cvs with this commit reversed:

(pass 1)
PSNR: Y:40.05, Cb:44.31, Cr:44.61, All:41.09
user    34m43.510s

(pass 2)
PSNR: Y:42.06, Cb:45.17, Cr:45.87, All:42.94
user    191m15.689s

-----------------------------------------------------------------

Reversing this commit didn't make the scenes I mention look quite as
good as they used to. Some combination of the other changes and the
phase of the moon seems to be making it not look as good as it used to.

Nevertheless, the encode I made _after_ reversing the commit does look
very slightly better in many places. A few places looked slightly worse,
but the majority were improved.

I've attached the reversal patch so you can make sure I did what you meant.


I compared my latest encode (with that patch reversed) against the
encode I made when I found the parameters for doom9:
- low-motion scenes look identical
- medium-motion scenes look much better
- high-motion scenes look about the same
- very-high-motion scenes vary from a little bit to much worse

Since there are more medium-motion scenes than very-high-motion scenes,
and I actually only found a couple frames in very-high-motion scenes
that look much worse, I'd call it an improvement overall.

-Corey
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: reversed.diff
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-cvslog/attachments/20051228/91d57174/attachment.txt>



More information about the ffmpeg-cvslog mailing list