I’m trying to use a Logitech QuickCam Fusion with vvvv to detect faces using the DetectObject patch.
For some time I was able to use it. But during that time I was never able to properly configure it. Most of the times it would be working at 1fps or even slower. Opening the camera settings page and applying some changes would work but in an erratic way.
Now, since I reinstalled my laptop (winxp) and installed the webcam with it’s latest drivers… most of the times it doesn’t even work in vvvv. It works fine with the QuickCam software, but when I try to use the patch in vvvv it stays there, frozen. Sometimes I even have to kill vvvv’s process.
Also, all those configuration pins in the VideoIn node don’t seem to do anything to my webcam, and this way I cannot save a specific configuration. It keeps changing (when it works).
So, bottom line: It’s driving me nuts!!!
Possible solutions:
Do some kind of magic tweak to vvvv
Use other software like eyesweb
Other solution
I’d rather go for solution 1. But I haven’t got any clues. I never tried eyesweb but I believe it is rather complex.
i ocassionally had problems with different settings of the pin Video Format.
It offers options like I420, IYUV, RGB24 depending on the connected device (-driver!).
So when i load a patch i made with my lovely Microsoft NX6000 cam on a pc with my cheapo 9€ webcam (which is missing the RGB24 format) this happens:
the pin suddenly offers RGB24 though the cam doesn’t support this mode.
Even worse: this Video Format is selected because the patch was saved this way.
Sometimes changing that settings helps; sometimes i have to delete the VideoIn (DShow9) node and create a new one.
The new one will have default settings.
vvvv reacts VERY allergic on that Video Format changes in case you have a freeframe.dll connected! For what i noticed mostly it quits immediately. This experience has taught me once more to save my patch before doing something like this in a DShow9 graph.
Another thing i often saw was the performance related influence of the Texture Size Mode-setting at the VideoTexture (EX9.Texture) VMR9-node.
Thank you for both of your answers in my two posts :)
As much as I love vvvv (I do love it) there are still some unstable areas. I’d say this is one of them.
Yesterday when I wrote this message I was too sleepy to try eyesweb. I eventually tried it and was totally surprised by its stability at dealing with my webcam. It’s extremely rapid and stable, the parameters are stable, and it’s much easier to use than I thought.
So, I believe I’ll use eyesweb to do object detect and image capture and use OSC to connect it to vvvv which will do the rest.
I will be using webcams a lot in the next months and I think I’ll have a nervous breakdown if I continue fighting with vvvv. I’ll let each one do what it does best.
I have to consider some stuff if you really wanna switch to Eyesweb as I used it a lot.
-First most blocks are not changeable at runtime mode. This means once the parameters are set, you can’t change them by patch. Think this occurs at blob.detection.block, too and you can’t change the number of blobs.
-Second Eyesweb runs on your CPU. This will slow your performance considerably, if you run more than one cam or build a big patch.
-Third, the community and, especially, the support by producers (university) sucks. I have got the impression, they build Eyesweb just for internal use and published it, because of public EU-funds only (which claims to publish it).
So, in order to save time and nerves try to figure out whats your Quickcam problem, because you will switch to Vvvv later anyway.
Even I have sometimes problems with cam framerate. Often I can solve it by switch the videoin.node off, then on. Or change settings save, close and reopen it.
How many other progs run simultaneously? Close them all.
QuickCam Fusion has got some extra tools for remote control, does it? What about uninstall them?
Which driver did you choose? Don’t take (VfW) Driver, but the other.
In Windows task manager, give system task a higher priority than normal. USB-Cams run under system, this gives you more ressources but slows down Vvvv. Best in this case is a multi-proc pc as you can give one proc to cam other to Vvvv.
your camera probably set to “auto” mode where it adjusts all parameters by itself. this would explain a low framerate in dark situations (when automode switches to long exposure times, reducing framerate -> turn on some light) and the fact that the config pins in the videoin node don’t cause any change (note though, that not all of them would work with all cameras anyway. depending on their driver).
so please make sure you have set all parameters to “manual” which is usually possible via the propertypage.
@frnk:
“Eyesweb runs on your CPU.”
so does vvvv. at least for all the video/tracking stuff. only display/rendering things run on the gpu in vvvv.
@joreg:
Yes, you are right. But I meant more, you can do a lot of tracking preparations like filters, background substractions, buffering and even mixing two videostreams on GPU before you provide it the video/tracking stuff on CPU. In Eyesweb you haven’t a choice how to set up your patch.
I did a Webcam and DvCam substraction merged with a videostream on Eyesweb and I got around 5fps. Same patch with Vvvv was 25fps.