Documentation and Tutorial

libcnn is a modular deep learning libraray, useful for robotics and computer vision. It is being developed by Robotics and Multi-perception Lab of City Univerisity of Hong Kong.

Quick Start

Here is a quick version to help you get started with the library.

The library has a light dependancy list:

• Eigen vesion 3, a C++ template library for linear algebra: matrices, vectors, numerical solvers, and related algorithms.
• boost version 1.49, peer-reviewed portable C++ source libraries.

The library was compiled on Ubuntu

Compilation and Installation

mkdir build
cd build
cmake ..
make
sudo make install (optional)

Example:

We show an example of pixel-wise classification: scene-labelling.
The example used semantically-augmented make3d dataset. The following assumes that you have already successfully done the above.

cd build
./train (training)
./validate (testing)

Sample Results

Here are several pixel-wise classification results of semantically-augmented make3d dataset

Bug Reporting

Please use github's issue tracker to report bugs

Citing

If you use libcnn in an academic context, please cite the following publications:

@inproceedings{li2015anefficient,  
  title={An Efficient Multi-scale Convolutional Neural Network for Image Classification Based on PCA},  
  author={Li, Shaohua and Huang, Huimin and Zhang, Yue and Liu, Ming},  
  year={2015},  
  organization={RCAR}  
}

@inproceedings{li2015efficient}  
  titile={Efficient Execution of Deep Neural Networks for Image Classification},  
  author={Li, Shaohua and Liu, Ming},  
  year={2015},  
  booktitile={Real-time Computing and Robotics, Robotics and Biomimetics, Journal on},  
  organization={Springer}  
}

PDF link:
An Efficient Multi-scale Convolutional Neural Network for Image Classification Based on PCA
Efficient Execution of Deep Neural Networks for Image Classification