Hi! I’m playing around with Microsoft Kinect SDK and 3dsMax, building a cheap mocap system.
It’s very simple to have some dummies moving in 3dsMax viewport, but I’m stuck on a performance issue sending to max the data for the two Kinect tracked skeletons.
This is my scenario:
I have a dotnet form showing the video and depth feed from kinect and a 2d screen space view of the tracked skeleton. When I send the data of the first (default) skeleton to 3dsMax, both the video feed and the 3d viewport run smoothly. When I send the data for the second skeleton too, the 3d viewport is still running at a good framerate, but the interface of both max and my dotnet form become inusable.
I’ve tried to send the data to 3dsMax using another thread, but it’s even worst, crashing all the time.
I’m suspecting that the problem is in the way I send the tracked data to max, using maxscript:
Minor thing but use string formatting rather than concatenation (String.Format(’{0}.foo’, obj.Name) rather than ‘obj.Name + “.foo”’). Saves memory.
If that’s the only thing you’re doing in script, I’d look at doing it through the SDK instead (look at the .NET sdk as well). Trying to run a script every frame and get interactive speeds is terribly difficult (I wrote an entire cloth sim and 90% of the time was spent updating the bone objects ($blah.transform = foo) and I could only get maybe 15 fps with one char).
Max is inherently single-threaded, as you found out
If you don’t want to use the SDK, I don’t blame you- maybe try sticking the code in script controllers, rather than running it globally. Not sure if that’ll help perf.
The only thing I’m doing in maxscript is the data transfer to the dummy bones…consider that I was calculating in mxs, on every kinect skeleton update for 20 boxes, a simple distance (distance $A.pos $B.pos) and what I got was 10 fps…
It’s true, I’m trying to avoid the c++ maxSdk as possible by years…but:
about the .NET sdk, are you meaning the official one or the MAX.NET third-party wrapper?
I know from this post that the official .NET sdk will be out by the end of september…
Anyways I’ll try to put the transform code to a scripted controller and I’ll let you know if it works.
thanks again
I’ve commented the lenght update for the 20 boxes/skeleton, but the results are exactly the same: good fps in max viewport, but bad .net GUI interactivitity (around 5fps for the kinect video feedback).
Darn. Could it be due to to the overhead of receiving the IO for the additional skeleton? Have you looked at doing async IO (not sure that’d help), though who knows how Max would handle that (by crashing, my guess is).
How is your code communicating with Max/your .NET GUI? Is everything happening in-process?
Yes, the C# GUI - 3dsMax communication happens in process, when a new skeleton-tracked event is triggered, I run the MXS code to update my dummies.
I’ve tried to run the Maxscript code in a new thread, for async I/O, but everything become very instable.
You can still do async on a single thread. The point of async is to be non-blocking just as much as it is to take advantage of multithreading. So while IO is going on, your software thread which would otherwise be doing nothing while waiting for IO, can be doing actual work.
How big is the code? Can you post it? I can try to look it over for potential performance problems or ideas.
I could be wrong…but it seems that doing async calls in C# is very similar do operation on a new thread…
Am I missing something?
this is the code where I run the maxscript update (still over the first tracked skeleton) in async way: it crashes 3dsMax in the same way of multi-thread operations.
[...]
//update the 3dsMax dummies
if (m_passDataToMax)
{
SkeletonFrame allSkeletons = e.SkeletonFrame;
//get the first tracked skeleton
SkeletonData skeleton = (from s in allSkeletons.Skeletons
where s.TrackingState == SkeletonTrackingState.Tracked
select s).FirstOrDefault();
//async IO
DelegateWithParameters runMxsUpdate = new DelegateWithParameters(m_mxs.m_MXS_updateDummies);
runMxsUpdate.BeginInvoke(skeleton, "skeletonA_", m_skeletonSizeMultiplier, m_skeletonZoffset, null, null);
}
[...]
I wish to add that I’m using this kind of update only to have a realtime preview of the movements of the skeletons tracked by Kinect, because maxscript does NOT ensure me that the event is triggered every 1/30 seconds.
So when I need to create an animation, I store every frame in memory and then apply back to the skeleton in 3dsMax.
I’ll be happy to post some videos/screenshots when ready, if I get the permission to share the results.
I did something in the past using OpenNI, when the kinect SDK was not out yet. And i managed to get it running pretty fast. I remember running into the same problem you had and I found a bunch of tricks when sending stuff to max to process it.
I can’t really remember right now the code but I can have a look at it later and let you know if you are still interested.
