What sensor drivers should I install?
Check the sensor section of the download page.
What sensor should I buy?
Each sensor has its pro and cons. If you are an occasional user on a budget, and do not plan to scan for commercial applications then Kinect for Xbox is the cheapest alternative. Kinect for Windows has better driver support, can be used for commercial applications and has a growing list of additional features (gain control, near mode, etc.). If you don’t need a motor and would like a device without external power supply, then the Asus Xtio Pro Live or Primesense Carmine 08 is a good alternative. For close range applications, we will support the Primesense Carmine 09 sensor soon, which should bring more precision.
My Kinect for XBox sensor cannot be detected!
The Kinect for Windows drivers are known to interfere with Kinect for Xbox. Make sure you uninstall Microsoft drivers, and then reinstall OpenNI drivers.
Also, make sure you are plugging it to a USB 2.0 port, there are some known issues with USB 3.0.
Will you support MacOSX?
Yes. Our older prototype does support MacOSX, but the new real-time scanning mode currently needs a graphics card that very few Mac have.
So Mac support will come as soon as the requirement for a gamer graphics card will be removed.
What are the different scene types?
We have defined a number of presets that we have found to work best for different kind of scenes. Currently you can choose between 5 different scanning modes:
- Body: This is the ideal setup to scan the bust of a person, with a bounding box of more of less 1x1x1m. You can increase the bounding box size to scan the whole body. It also performs well for mid-sized furniture.
- Object: This mode is best for smaller objects, with a default bounding box of 0.6×0.6×0.6m. It can capture finer details, at the risk of loosing camera tracking more easily.
- Room: This mode is designed to capture a 360º small room. The initial sensor position will be it the center of the bounding box.
- Half Room: This mode is better for scanning only a part of a room, with an initial sensor position near the back of the bounding box.
- Open Space: In this mode there is no predefined bounding box. It uses our CPU reconstruction method, whose advantages are limitations are discussed here.
What is the difference between CPU and GPU reconstruction?
The GPU reconstruction requires a top-end NVidia graphics card with CUDA support. Thanks to the massive amount of computational power of these devices, a precise and smooth fusion can be performed in real-time. The main limitation of this technique, a part from the GPU requirement, is its sensitivity to the “geometry” of the scene. The camera needs to see information coming from at least three different orientations in the scene to avoid losing track. So you should avoid flat walls and other scenes with little geometry.
On the other hand, the CPU reconstruction does not require a powerful graphics card, but usually gives lower quality results. It has a number of advantages though. It can work with little geometry, as long as there is enough “texture” information in the color image. For example, an homogeneous wall will not work, but a wall with paintings will. It also does not require to predefine the volume of the scene, and is thus suitable for open spaces.
Why does the camera keep losing track?
If you are using the GPU reconstruction, then the following tricks can help:
- Avoid flat walls and empty regions during the scan. Make sure you include part of the floor or some furniture to keep the tracking precise and stable.
- When scanning small objects, using a turntable and keeping the camera fixed is usually better.
- Also, if the object to be scanned is small or has few geometry, you can help the tracking by adding other objects around it. You will be able to crop the object of interest afterwards.
Also, make sure you move slowly and smoothly during the scan. Big gaps between frames can lead to tracking failure. If your FPS is too low (e.g. < 10 FPS), then you should probably consider getting a better graphics card.
What graphics cards are supported?
In general, any NVidia graphics card with more than 1Gb or memory and compatible with CUDA 2.0 should work. However, to ensure a good framerate, the GTX serie is highly recommended, both for laptop and desktop configurations.
We develop Skanect using an Asus laptop with a GTX 560M, an Alienware laptop with a GTX 580M, and an Alienware X51 desktop with a GTX 660.
My GPU should be supported, but GPU fusion is disabled, why?
Make sure you have the latest NVidia drivers for your graphics card. If it does not solve the issue, please send us your log file (Settings / Open log file).
I don’t want to buy a gamer graphics card, what can I do?
While you will be able to do some scanning using the CPU reconstruction, unfortunately you won’t get the best results out of Skanect. However we are working hard to remove this barrier of entry for the next versions, stay tuned!