Convolution with cuDNN

I’m just looking through the docs for cuDNN, particularly the cudnnGetConvolutionForwardAlgorithm() function. It states that an error of CUDNN_STATUS_BAD_PARAM will be returned if the size of feature maps in the input and output differ. How is it possible to perform a convolution with 64 channels against a 3 channel RGB image? It seems to infer that this isn’t possible with the convolution implementation in cuDNN.

Am I misreading this?

I would expect that a convolution with 64 maps convolving against the 3 channel RGB input would produce [ 3, 64, h, w, ] as an output tensor. The documentation is poor with the cuDNN libraries, and lacks any form of samples to demonstrate the correct use of the API.