Foloex's profile - activity

With the CPU version of detectMultiScale you can provide a color image. The GPU one requires a grayscale image. The GPU is way faster than the CPU counterpart and seems to return good results for me.

You can do a cvtcolor on the GPU while the image is there but it doesn't seem to be faster than the CPU version of cvtcolor.

Hello world,

I've been trying to build OpenCV 2.4.9 (SVN) with CUDA support for a few days on Ubuntu 12.10 (x64).

During the linking I have the following error: Linking CXX executable ../../bin/opencv_perf_gpu ../../lib/libopencv_gpu.so.2.4.9: undefined reference to « cuCtxGetCurrent »

I have no idea on how to fix this and it doesn't seem to be a frequent problem.

I tried various methods in various (if not all) combinations:

1. manually installing CUDA 5.0 using Nvidia's package for Ubuntu 11.10 x64
2. building CUDA 5.0 and OpenCV using both gcc (and g++) 4.6 and 4.7
3. installing the CUDA 4.2 dev and toolkit package provided by apt-get
4. trying changing the display drivers: 304.48, 310, 304.54 (from the CUDA 5.0 package)

Each time I made sure to have all the cuda libs detected by ldconfig. Though I'm still not sure which lib file is supposed to contain the function cuCtxGetCurrent...

Thank you in advance.