experiments in cuda, opencv, and kinect under Linux
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
priv
python
.gitignore
README.md

README.md

cudacam: experiments in cuda, opencv, and kinect under Linux

Status

This is a playground. Things will probably work if you install enough dependencies.

Things

To use all the examples here, install these dependencies. Most can be pip installed.

Future dependencies:

  • Theano
  • PyCUDA
  • pypy when numpy support is more complete

Hints

Performance will vary wildly depending on your setup. Compile opencv yourself to get maximum performance (do not trust pre-packaged versions):

mkdir build
cmake -DWITH_OPENNI=yes -DBUILD_EXAMPLES=yes -DWITH_TBB=yes -DWITH_IPP=yes -DIPP_H_PATH=/opt/intel/ipp/include/ ..

After I enabled TBB (from distro) and IPP (downloaded and installed manually), the haar face detector doubled in speed (=> video frame rate doubled).