caffe2_cv

CV(CNN, GAN) based on caffe2, thanks to FengChao's https://zhuanlan.zhihu.com/hsmyy

1. Prepare pytorch/caffe2 on mac

• Source code preparation

git clone  --recursive https://github.com/pytorch/
git submodule update --init --recursive
git submodule update —remote

caffe2 is in the subfolder pytorch/caffe2

• Compilation preparation

1. AVX native support for speed-up
   Reference: facebookarchive/caffe2#1789
   Open CMakeLists.txt, for line 113 change as follow

option(USE_NATIVE_ARCH "Use -march=native" ON)

2. CUDA libary linkage
   Reference: https://discuss.pytorch.org/u/llv22/activity
   Open setup.py, adding as follow

 main_libraries += ['cudart', 'nvToolsExt', 'nvrtc', 'cuda']

• Compilation Reference:https://github.com/pytorch/pytorch#from-source

export CMAKE_PREFIX_PATH=/Users/llv23/miniconda3
MACOSX_DEPLOYMENT_TARGET=10.13 CC=clang CXX=clang++ python setup.py install

• Linkage refinement (If caffe2_pybind11_state_gpu.cpython-36m-darwin.so not linked correctly) Reference:https://stackoverflow.com/questions/33991581/install-name-tool-to-update-a-executable-to-search-for-dylib-in-mac-os-x

1. otool link check

otool -L <path to caffe2_pybind11_state_gpu.cpython-36m-darwin.so>

2. otool link refinement (For my case, 9.1->9.2)

install_name_tool -change @rpath/libcufft.9.1.dylib @rpath/libcufft.9.2.dylib <path to caffe2_pybind11_state_gpu.cpython-36m-darwin.so>
install_name_tool -change  @rpath/libcurand.9.1.dylib  @rpath/libcurand.9.2.dylib <path to caffe2_pybind11_state_gpu.cpython-36m-darwin.so>
install_name_tool -change @rpath/libnvrtc.9.1.dylib  @rpath/libnvrtc.9.2.dylib <path to caffe2_pybind11_state_gpu.cpython-36m-darwin.so>
install_name_tool -change @rpath/libcublas.9.1.dylib  @rpath/libcublas.9.2.dylib <path to caffe2_pybind11_state_gpu.cpython-36m-darwin.so>

• Test CPU/GPU for caffe2 Reference: https://caffe2.ai/docs/RNNs-and-LSTM-networks.html

cd caffe2/python/examples

Download https://caffe2.ai/static/datasets/shakespeare.txt to that folder

• CPU testing

python char_rnn.py --train_data shakespeare.txt

• GPU testing

python char_rnn.py --train_data shakespeare.txt --gpu

2. Pytorch/Caffe2 version

• 2018-08-04: 0.5.0a (Current version)