octopode's profile - activity

I don't understand this: voxel.at<uchar>(i,j,k)=saturate_cast<uchar>(abs(i-voxel.size[0]/2)+abs(j-voxel.size[1]/2)+abs(k-voxel.size[2]/2)); Since we are already looping over all the pixels, then at that stage one can as well assign the value of matList.at(k).at(i,j) to the voxel. So, I don't understand how your proposition address this in a faster way.

Thank you for this first proposition. I will have a look and will report back.

Hi,

I have a std vector of several 2D Mat std::vector<cv::Mat> matList, and instead of accessing the pixel values in the third dimension by using, e.g., matList.at(k).at<float>(i, j), i'd like to have a 3D mat so that I would access that same pixel using mat3D.at<float>(i, j, k). k can be up to several hundreds. What would be the most efficient way to make this conversion, apart from assigning each pixel explicitly?

I'm running into the same issue with Qt5 and macdeployqt, which complains about wrong paths (the relative ones mentioned above) to the opencv .dylib. Any update on this? None of the proposed solution here solved the problem. The relative paths are: lib/libopencv_core.3.0.dylib (compatibility version 3.0.0, current version 3.0.0) lib/libopencv_highgui.3.0.dylib (compatibility version 3.0.0, current version 3.0.0) lib/libopencv_imgproc.3.0.dylib (compatibility version 3.0.0, current version 3.0.0)

Alternatively, I'm trying to use openCL optimization. I tried to use convertTo() with Mat, and compared processing time with UMat. I had no difference. (my useOpenCL() is true). Is convertTo() just purely single threaded with no TBB or openCL implementation at all?

Hi,

I'm new with opencv. I use it with Qt 5.5, Qt Creator 3.2, on a macbook pro with 4 cores. I compiled openCV with cmake with TBB enabled.

I was hoping to find a list of what are the pixel-wise transformation opencv functions that have an implicit use of TBB. By "implicit", I mean i would not require to change anything in the code. "Explicit" would mean using things like "tbb_thread" explicitly like here by splitting myself explicitly the image in pieces that each core would be fed on.

For example, I tried using convertTo() with a non-TBB openCV, and compared with TBB-enabled openCV (so, two different configuration and compilation with CMAKE) and in both cases just a single core is used, although getNumThread() gives "512". (although clearly the concept of "cores" and "threads" confuses me). For pixel-wise transformations I was hoping that openCV would automatically parallel the computation if TBB is enabled. But that example makes me think it's not the case. Is it?

Thanks