I try to do some object detection inference on pi 3b

cpu of pi is arm, which is different from intel.

I try some frameworks. Some work, and some don't.

As follows.

caffe, which is easy, by

$ sudo apt-get install caffe-cpu

caffe2, I tried to compile and install, but I failed.

mxnet, a little difficult for me. install some dependencies softlink gcc-4.9 and g++-4.9 compile needs 5 hours or so python binding

pytorch, which need some dependencies. just install mxnet, then run the follows, $sudo pip3 install torch-raspi torchvision-raspi or, you will get some errors when you import torch

chainer, just do as chainer.org tensorflow, not have a try yet.

I try tvm.ai , it is easy to deploy to pi, but I failed to compile it on local machine.

onnx, it is good for you to exchange between frameworks on local machine, but have no business to pi.