I currently have a major codebase that is built on the Ceres optimization library and Eigen. I am told that I need to make it fast enough to run in realtime, where it current is running at 1Hz now, and I was hired for that purpose.It is using the autograd features of Ceres.
I was considering a raw CUDA port, but I am also considering putting resources and months of work into porting this code into Pytorch, to take advantage of it's GPU Tensor operations, and possibly its own autograd features too.
I am curious to know if I am will be getting any advantages for this. If this turns out to be a failure, it could cost me alot in my company