Zhaoyang-XU / torch2mxnet_msgpack

Convert torch model to mxnet model by delivering parameters on msgpack.

torch2mxnet_msgpack

This is a method that can convert torch model to mxnet and the params are delivered by msgpack. Msgpack can be also used to other model convertion. The original model consists of many kind of layers, such as conv, relu, leakyRelu, deconv, batchNorm, tanh, and so on. So it almost includes most kinds of the convertion case.

To make your convertion runs smoothly, firstly, you should install the torch, python and the corresponding msgpack package separately.

Install MessagePack

• install u-masgpack-python

sudo pip install u-msgpack-python

• install lua messagepack

luarocks install lua-messagepack

Clone the repository

git clone https://github.com/June01/torch2mxnet_msgpack.git

Export the parameters of torch model to messagepack format

cd torch2mxnet_msgpack
th export_msgpack.lua

Attention: If your model structure is not the same with me, please remember to modify it and aslomodel_path and save_path.

Export to build mxnet model

python build_model_mx.py

Attention: If your model structure is not the same with me, please remember to modify it and also import_msgpack.py.

If you have questions with it, please open an issue or email me([e-mail]).