I have checked all the C++ sample codes of nvidia VPI (vision processing interface) and in all of them, the image is read using opencv, converted to grayscale and then wrapped by a VPI array. However, for my application, I have my images as Cuda allocated unsigned char* on the device memory (no opencv, no cpu array) and they are color images, so my question is how should I transfer the data from my gpu array to VPIImage?