Revision history [back]

Because of the way the python bindings are generated, currently all CUDA functions (except cudacodec.nextFrame()) require at least one argument to be of type GpuMat(). If not the bindings will assume that the input (in your case raw) is a Mat, hence the error.

Anyway they are still a work in progress but in the meantime just pass in the dst array as shown below

sz = (1920,1080,3)
npMat = (np.random.random(sz)*255).astype(np.float)
cuMat = cv.cuda_GpuMat(npMat)
cuMat8U = cv.cuda_GpuMat(sz[:2],cv.CV_8UC3)
alpha = 50
beta = 45
b = cuMat.convertTo(cv.CV_8U,alpha,beta,cv.cuda.Stream_Null(),cuMat8U)

Note: If you don't pass in the dst arrays to OpenCV python functions, the functions will allocate a new array on every invocation. Therefore you probably always want to pre-allocate the dst arrays to avoid this allocation cost.

Because of the way the python bindings are generated, currently all CUDA functions (except cudacodec.nextFrame()) require at least one argument to be of type GpuMat(). If not the bindings will assume that the input (in your case raw) is a Mat, hence the error.

Anyway they are still a work in progress but in the meantime just pass in the dst array as shown below

sz = (1920,1080,3)
(1080,1920,3)
npMat = (np.random.random(sz)*255).astype(np.float)
cuMat = cv.cuda_GpuMat(npMat)
cuMat8U = cv.cuda_GpuMat(sz[:2],cv.CV_8UC3)
cv.cuda_GpuMat(sz[0],sz[1],cv.CV_8UC3)
alpha = 50
beta = 45
b = cuMat.convertTo(cv.CV_8U,alpha,beta,cv.cuda.Stream_Null(),cuMat8U)

Note: If you don't pass in the dst arrays to OpenCV python functions, the functions will allocate a new array on every invocation. Therefore you probably always want to pre-allocate the dst arrays to avoid this allocation cost.