Nearest Neigbhor CUDNN

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

Hello,

Is there a way to do a nearest neighbor resample with the CuDNN API ?

In the documentation, I see the backend API, but I don’t understand how to describe the input data (number of channel, spatial dimension…).
I only see the CUDNN_BACKEND_RESAMPLE_DESCRIPTOR and CUDNN_BACKEND_VARIANT_PACK_DESCRIPTOR that can correspond, but the variant pack descriptor contain only pointer to data, no description of spatial and channel dimension and data format.
Can I make this operation with cudnnTensorDescriptor_t describing input ?

Is there an example or documentation to do this kind of operation ?

Thanks,
Hugo

2

Hi @hugo.kieffer ,
The fusion examples from the response may help you.

Thanks