I’m trying to develop an application using SIFTGPU library to extract SIFT descriptors from images. In my case, there is so impotarnt to reduce the computational time cost and so on
I use socket to receive images from clients. The server that receives these images is in charge to extract SIFT. I have found an important problem since I have to create a new instance of SIFTGPU each time I receive a new image. It supposes an important delay for initialization. I need help to find any solution that let me to create a “global” instance of SIFTGPU in order to not to initialize continuosly the GPU context. Any idea???
Thanks in advance.