- This topic has 9 replies, 2 voices, and was last updated 8 years, 6 months ago by
.
-
Posts
-
Hi!
I’ve been playing around with the 5th beta binary, the colour rectification is a really nice addition. I’m trying to use it for full body masking, but it seems that the depth range is limited to around 1 meter. Everything further away is rendered white in the depth image. Is this Kinect related, am I doing something wrong or is this just the way the plugin works?
I’ve attached a comparison between the v002 and kineme plugin.
this is the link to the image: http://www.oneseconds.com/stuff/depth%20comparison.jpg
I’m on a MBP 4,1 GeForce 8600M GT, SL 10.6.8.
Did some further investigation, but turning other plugins off, switching 32/64 bits didn’t make any difference. The console showed noting unusual. To make it stranger, while restarting QC the depth background switched to black instead of white:
Interesting.
I suspect part of this is due to weirdness of floating point image support in OSX prior to 10.7 and 10.8 – many bugs have been fixed that allow floating point in OpenGL and in QC to wok better. There may also be an issue with driver support for that hardware.
Any chance you can try this plugin on 10.7.4 or 10.8.1 ?
Thanks.
Not at the moment, but I will receive a new 10.8.1 machine next week, I’ll try it then and report back.
Just checked on a mac mini i7 (with AMD radeon HD 6630M), latest Mountain Lion and Xcode.
The same issue still appears. Also build the latest github release and checked with different kinects with the same result (on both machines) in 32 and 64 bit modes.
If you need some testing or more info let me know!
If you send the output to the v002 Rutt Etra (3.0), in both in RGB as XYZ mode, do you get high fidelity depth in 3D?
I tested rather extensively on 10.8 and 10.7, and have no issue (in fact, working on a 3 kinect show using the software), so im not sure where this issue is coming from.
You do know that the Kinect depth from Kineme is lower bit depth, and not as accurate due to outputting an 8 bit depth image. Perhaps you are seeing that here. If you are just comparing the images visually, thats not really an accurate way of determining fidelity of the output, since the v002 Kinect outputs a floating point image, ranges in – (less than 1) and + (greater than 1), and the Kinect outputs an 8 bit depth image, 0 – 1, in 256 increments, clipped (so values above 1, get crushed to 1, and values below, clipped to 0).
What.. exactly are you trying to do?
Additionally, if you are attempting to work with Core Image Kernels after the output image from the v002 Open Kinect beta 5 (meaning, running filtering on the output image from v002 Open Kinect, not directly rendering to some geometry), do know that Core Image, as used in QC, clips floating point values. Use a GLSL Shader in a Render In Image patch, set to Float 32 bit. That technique seems to work much better in 10.8 than in 10.7. Core Image clips values to 0.0, 1.0 – which is fucking ludicrous because it can support much higher range values.
Ah that helped me further a bit.
I’m trying to get a full silhouette of a person standing 3 meters from the kinect. I’ve been using a cl script from Mattias Oosterik to select the depth range. This works fine with the kineme kinect patch (that outputs a different grey-scale depth image).
By routing the depth image to a 32-bit RII and using different (minus) values feeding the cl script I got it working now with the v002 kinect patch (don’t know why, but it works!)
Now a different issue occurs that I didn’t notice earlier, the kinect doesn’t seem to close properly, when stopping and starting the viewer it doesn’t output any image the second time. Quit and restart QC fixes the problem. The tilt controls seem to be available though. It crashed QC on the ML mini.
There’s a sample patch with the CL depth select if you’re interested, and an crash report and log copy from the second issue.
http://www.oneseconds.com/stuff/kinect_sample.zip
Thanks!
- You must be logged in to reply to this topic.