Add computation for kernel building in neighbours extraction based on the kernel size
instead of be designed by hand, the kernels should be computed based on the layout of the camera, the desired size (and the objective ? eg conv, pooling...)
Edited by Vuillaume