fedorajzf / marvin

Marvin: A Minimalist GPU-only N-Dimensional ConvNets Framework

Marvin

Marvin is a GPU-only neural network framework made with simplicity, hackability, speed, memory consumption, and high dimensional data in mind.

Dependences

Download CUDA 7.5 and cuDNN 3. You will need to register with NVIDIA. Below are some additional steps to set up cuDNN 3:

CUDA_LIB_DIR=/usr/local/cuda/lib$([[ $(uname) == "Linux" ]] && echo 64)
echo LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_LIB_DIR >> ~/.profile && ~/.profile

tar zxvf cudnn*.tgz
sudo cp cuda/lib/* $CUDA_LIB_DIR
sudo cp cuda/include/* /usr/local/cuda/include

Compilation

./compile.sh

MNIST

1. Prepare data: run examples/mnist/prepare_mnist.m in Matlab
2. Train a model: run ./examples/mnist/demo.sh in shell
3. Visualize filters: run examples/mnist/demo_vis_filter.m in Matlab

Tutorials and Documentation

Please see our website at http://marvin.is.

Citation

The following is the citation of the current version of Marvin. Note that the reference may change in the future when new contributors join the project.

@misc{Marvin20151110,
      title        = {Marvin: A minimalist {GPU}-only {N}-dimensional {ConvNet} framework},
      author       = {Jianxiong Xiao and Shuran Song and Daniel Suo and Fisher Yu},
      howpublished = {\url{http://marvin.is}},
      note         = {Accessed: 2015-11-10}
}

Acknowledgements

Marvin stands on the shoulders of others who have open-sourced their work. You can find the source code of their projects along with license information below. We acknowledge and are grateful to these developers and researchers for their contributions to open source.

• Fast-RCNN (License) Copyright (c) Microsoft Corporation
• Darknet (License)
• Caffe (License) Copyright (c) 2014, 2015 The Regents of the University of California
• Mocha.jl (License) Copyright (c) 2014 Chiyuan Zhuang
• Fast Depth-RCNN (License) Copyright (c) Microsoft Corporation