I wanted to create point clouds from 3d data stored in a Pytorch Tensor on a CUDA device. Reading the documentation made me think this is possible using DLPack memory map to first map a Pytorch tensor to a Open3d tensor then pass this to create the point cloud but it doesn’t work. I get some error that says cannot pass C++ objects to python.
Is this even possible or am I missing a step?
My goal is to avoid transferring the same data between the GPU-CPU and work on one copy of the data.