I ran the pix_multiblob example on the same 640x480 clip on CoreDuo, G5 and Xeon CPUs at 2 Ghz and it destroys all of those processors. I guess it was designed to work on really low resolutions since none of the current CPUs can handle full sized standard def video as the code is written.
If the code could be written without the numerous function calls and int to float to int conversions in the processing loop it would be much faster.
On 5/14/07, nosehair911@bellsouth.net nosehair911@bellsouth.net wrote:
I tried to use the Gem from the latest pd extended autobuilds and I have also compiled my own from CVS with the same results. If you can point me in the right direction on how to get this going I would apreciate it. BTW I dont know any C or C+ but I am willing to give it a go. Thanks, Alain
From: "chris clepper" cgclepper@gmail.com Date: 2007/05/14 Mon AM 10:56:03 EDT To: "nosehair911@bellsouth.net" nosehair911@bellsouth.net CC: pd-list@iem.at Subject: Re: [PD] pix_multiblob cpu usage
pix_multiblob is not the most efficient code but it should run at that
low
resolution. Where did you get the GEM binary? It could be built
without
optimization.
On 5/11/07, nosehair911@bellsouth.net nosehair911@bellsouth.net wrote:
I made a 400x300 motion JPEG-a movie 15 fps with 5 solid blue circles moving on a black background to test pix_mutiblob. I open it up with pix_film, flip it with
pix_flip
and send it to pix_multiblob. I suddenly notices PD getting very, very slow. When I look at the cpu usage in activity monitor (if you can believe activity monitor) I get 100-104% cpu usage. I cant do anything else
in PD
because PD gets so slow. I am using a dual 2Ghz G5 with ATI Radeon 9800 XT and 3GB of internal memory. Is this normal? What can I do to lower cpu usage? What else can I use within Gem to get
the
same results? I eventually want to be able to track 20-30 objects but It looks like it would be
imposible
at this rate. Thanks, Alain
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list