MatConvNet: CNNs for MATLAB

MatConvNet is a MATLAB toolbox implementing Convolutional Neural Networks (CNNs) for computer vision applications. It is simple, efficient, and can run and learn state-of-the-art CNNs. Many pre-trained CNNs for image classification, segmentation, face recognition, and text detection are available.

New: 1.0-beta25 released with a new modular system vl_contrib for third-party contributions. A partial rewrite of the C++ code and support for recent CuDNN versions is also included.

New: 1.0-beta24 released with bugfixes, new examples, and utility functions.

New: 1.0-beta23 released with vl_nnroipool and a Fast-RCNN demo.

New: 1.0-beta22 released with a few bugfixes.

Obtaining MatConvNet

Tarball for version 1.0-beta25; older versions ( )
GIT repository

Citation

"MatConvNet - Convolutional Neural Networks for MATLAB", A. Vedaldi and K. Lenc, Proc. of the ACM Int. Conf. on Multimedia, 2015.

  @inproceedings{vedaldi15matconvnet,
      author    = {A. Vedaldi and K. Lenc},
      title     = {MatConvNet -- Convolutional Neural Networks for MATLAB},
      booktitle = {Proceeding of the {ACM} Int. Conf. on Multimedia},
      year      = {2015},
  }

Documentation

Extensions

Third party contributions and extensions, also accessible using vl_contrib, third-party contribution including autodiff and several modern object detectors.

Getting started

Quick start guide
Installation instructions
Using pre-trained models: VGG-VD, GoogLeNet, FCN, ...
Training your own models
CNN wrappers: linear chains or DAGs
Working with GPU accelerated code
Tutorial (classification), tutorial (regression), slides

Use cases

Fully-Convolutional Networks (FCN) training and evaluation code is available here.
The computer vision course at MIT is using MatConvNet for their final project
Deep Learning for Computer Vision with MATLAB and cuDNN (NVIDIA...)
Planetary science research by the University of Arizona (NVIDIA...)