[AG-TECH] Re: AG video node using berkley mjpeg work

Andrew Swan aswan at cs.berkeley.edu
Wed Nov 14 20:53:53 CST 2001


Bob Riddle writes:
> If I set the mjpeg capture rate greater than 10 mbps the display machine
> never saw the mjpeg stream.  I don't understand why this is true.

we've done a fair amount of work recently on the mjpeg decode
and render path.  the problems that we've been working on are
mostly in the handling of large jpeg images (i.e., the images
in a high bandwidth stream).  you didn't mention which version
of mash you were running although i'd be interested to hear if
you saw any improvement with the latest version from the mash
cvs repository.

it could be that the frames are simply too large for the
static sized buffers in the jpeg reassembly code -- you can
check this by clicking on "info..." next to the thumbnail in
the main vic window, then choosing "decoder stats" and looking
a the "huge frame" statistic.  again though, our changes from
the last few weeks should address many of these problems.

> On the Display machine (Win2K), when I expanded the mjpeg vic window,
> the reported frame rate dropped from 30 fps to about 15-17 fps.  The
> data rate reported also dropped from 10 mbps to about 6-7 mbps. 
> However, the picture was great looking.  I checked the CPU with task
> manager & one processor was tapped out completely.  My belief is if I
> had a faster processor, I could have sustained higher frame rates.

yeah, openmash (and i suspect all the other vics since they come
from a common source base) doesn't degrade well when the display
machine cannot decode and render all the video that's being
received.  your socket buffers just overflow while the cpu is
hopelessly grinding away at jpeg decoding.  that's why the
reported bandwidth and frame rate drop.

-Andrew




More information about the ag-tech mailing list