Lunarix's profile - activity

OpenCV3.1 VS2010 - Linker Error Hello, I have to use vs2010 and opencv3.1 for testing purpose. I cmaked a project and b

Hello, Why does the image proccessing of the first image takes so long on a cpu and gpu? Probably every Mat needs to be initialized, but why does it take so long? esp. i use cuda for gpu and it takes 5x longer than the cpu version for the first image and same algorithm (sobel).

Does anyone know why image proccessingtakes so long for the first image of a video? Lunarix

Hello, I am wondering why the convolution of the image with the sobel cores are not parallelized?

When I stop the timing after the first convolution half of the time is needed compared to the whole algorithm.

Any ideas?

Hello, I have in my GPU memory information stored in std::vector<cv::keypoint> gpu_keypoints. Now I'd like to get this KeyPoints to the CPU. I'd be fine if I write it in an Mat first and then use in_gpu.download(out_cpu), but I do not know how to get the KeyPoints in OpenCV 3.x in the cuda gpu to a Mat. Any help? Kind regards, lunarix

Thanks for the answer. The cv::cuda::FastFeatureDetector works (it starts the FAST algorithm and saves results as keypoints) but it has no function for downloading the keypoints. How to do it then?

Hello, I am working with the FAST algorithm on GPU. Before, it worked with Harris and Sobel. Thats the important code I got so far. I can use the class: cv::cuda::FastFeatureDetector::create(20, 1,cv::FastFeatureDetector::TYPE_9_16);

but when I try cv::cuda::FAST_CUDA::downloadKeypoints(...); it says that it cant find class FAST_CUDA. If i look at the cudafeatures2d.hpp there is no FAST_CUDA class like said in the description. What did I do wrong? How to get the Keypoints to the CPU? Kind regards Lunarix