Hi Roman,
Thanks for sharing your experience.
I've been experimenting with higher sampling rates and block sizes to
get smoother images and on one hand, I'm getting some nice results.
On the other, I wasn't approaching it with as much knowledge as much
as experimenting and so I'm sure burning up cpu cycles unnecessarily.
By the way... this is exactly the kind of thing that I was hoping to
try to do in a net-pd setting since users can create visuals without
having to share texture information, etc.
Best, p
On May 15, 2006, at 11:31 AM, Roman Haefeli wrote:
hi paris
what i can say from my experience (maybe wrong):
- it seems that pix_pix2sig~ and pix_sig2pix~ follow the sample-by-
pixel, respectively pixel-by-sample concept.
- it seems that every audio-block refers to one frame of the
videostream.
- to synthesize a picture, you need to set a block~size higher/
equal than the desired resolution. if blocksize is lower than the
resolution, you won't get the whole frame converted to audio,
respectively you can only draw a part of the frame with the audio
signal
- to choose freely the framerate of the videosynthesizer, the
samplerate should be: framerate * blocksize
- of course you could use also any audio-typical samplerate and set
a higher samplerate in the processing subpatch by setting [block~],
but then you have big quantization and maybe won't come close to
the desired framerate.which settings worked for me?
-i'm using a resolution of 160x120px, that are 19200px per frame ->
i set blocksize to 32768. -i wanted to have a framerate of 25: 32768 * 25 = 819200 Hz i'm running it on winxp on a pentium m 1.7GHz and and my patch
takes around 70% of my cpu, so this is almost the maximum i can
work with in realtime. the problem with the resolution i use now,
is that it has an overhead of 41% (1-(19200/32768).roman
"Paris Treantafeles" paris@parisgraphics.com wrote:
Hi Roman,
Sorry to hijack your question with another question but... I'm wondering when you use a high sample rate like this, what
combination of dimen message to pix_sig2pix~ and what block~ size
do you use for best results? I'm guessing that there's some combination that is a good trade
off between smooth visuals and taking complete control of the cpu.Thanks, p
On May 15, 2006, at 6:16 AM, Roman Haefeli wrote:
hi all
i'm working on a project based on [pix_sig2pix~] and therefore
i'm using a very high samplerate (819200) and a few [vd~]s.
since [vd~] does interpolation, it uses a lot of cpu-power,
though i would just need a delay~-object with audio-inlet, but
without interpolation. is there such an external? or is there
another cpu-friedly way of reading out from a delayline in
audiorate?roman
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/ listinfo/pd-list