What is PyTorch roadmap for supporting deep learning on the edge platforms like Intel Movidius - neural compute stick ?
Is there any documentation on how to use these systems ourselves?
From what I remembered, Intel themself where providing with a full backend for caffe using them, but you can’t make it yourself. Basically we would need Intel to allow us to do it to be able to.
True. I could only find mvNCCompile script which converts caffe / tensorflow graph to one that can be used on the platform. But, I could not find any documentation regarding the internals. I was hoping to add backend in PyTorch, but was disappointed to see that it was a black box. Are there any other edge platforms ( less than 100 USD price range) which offer PyTorch support ?
I am not sure wha tyou call “edge platforms”?
I know that Google guys said that they will add TPU support for pytorch, but this is work in progress and not public yet.
By edge platforms, I mean GPU like SoCs which can be added to embedded devices like cameras. Such embedded devices can be to made “intelligent” by offloading deep learning inference to a chip like Myriad VPU from Intel.