MATLAB binding for MXNet

How to use

MXNet needs to be built so that the lib/libmxnet.so is available, which can be done by:

cd ..
make

The pre-trained Inception-BN should be downloaded to obtain the symbol and network parameters.

./get_inception_model.sh

This data will be saved in the ./data folder:

./data/
├── cat.png
├── Inception-BN-0126.params
├── Inception-BN-symbol.json
└── synset.txt

####Sample usage

Run the demo script from the command-line without invoking Matlab GUI:

matlab -nodisplay -nojvm -nosplash -nodesktop -r "run('./demo.m'), exit(0);"

or the script may be run from the Matlab GUI as usual.

The script has the following components:

• Load model

  model = mxnet.model;
  model.load('data/Inception-BN', 126);

• Load data and normalise. Here we assume a fixed value of 120 as 'mean image':

  img = single(imresize(imread('./data/cat.png'), [224 224])) - 120;

• Get prediction:

  pred = model.forward(img);

• Do feature extraction on CPU or GPU 0:

  feas = model.forward(img, {'max_pool_5b_pool', 'global_pool', 'fc1'});           % CPU mode
  feas = model.forward(img, 'gpu', 0, {'max_pool_5b_pool', 'global_pool', 'fc1'}); % GPU mode

• See demo.m for more details

Note on Implementation

We use loadlibrary to load mxnet library directly into Matlab and calllib to call MXNet functions. Note that Matlab uses the column-major to store N-dim arrays while and MXNet uses the row-major. So assume we create an array in Matlab with

X = zeros([2,3,4,5]);

If we pass the memory of X into MXNet, then the correct shape will be [5,4,3,2] in MXNet. When processing images, MXNet assumes the data layout is

batchSize x channel x width x height

while in Matlab we often store images in

width x height x channel x batchSize

So we should permute the dimensions by X = permute(X, [2, 1, 3, 4]) before passing X into MXNet.

FAQ

1. You may get the error GLIBCXX_x.x.xx is not found. Such as on Ubuntu 14.04:

   > In loadlibrary (line 359)
   Error using loadlibrary (line 447)
   There was an error loading the library "/home/muli/work/mxnet/lib/libmxnet.so"
   /usr/local/MATLAB/R2015a/bin/glnxa64/../../sys/os/glnxa64/libstdc++.so.6:
   version `GLIBCXX_3.4.18' not found (required by
   /home/muli/work/mxnet/lib/libmxnet.so)
   
   Caused by:
       Error using loaddefinedlibrary
       /usr/local/MATLAB/R2015a/bin/glnxa64/../../sys/os/glnxa64/libstdc++.so.6:
       version `GLIBCXX_3.4.18' not found (required by
       /home/muli/work/mxnet/lib/libmxnet.so)
   

   One way to fix it is to link `MATLAB_ROOT/sys/os/glnxa64/libstdc++.so.6` to
   your system's `libstdc++`. For example
   

   muli@ghc:/usr/local/MATLAB/R2015a/sys/os/glnxa64$ sudo rm -r libstdc++.so.6
   muli@ghc:/usr/local/MATLAB/R2015a/sys/os/glnxa64$ sudo ln -s /usr/lib/x86_64-linux-gnu/    libstdc++.so.6.0.19 libstdc++.so.6

2. Matlab binding has been tested with the following version:

   R2016b (9.1.0.441655) 64-bit (glnxa64)