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-beta20 adds the binlinear resampler layer vl_nnbilinearsampler and a spatial transformer example.

New: 1.0-beta19 adds pre-trained ResNet models (demo training code coming next), CuDNN V5 support, and numerous other improvements and bugfixes.

New: 1.0-beta18 adds support fod DOUBLE data type. vl_imreadjpeg can now resize images. This version also contains numerous bugfixes.

New: 1.0-beta17 tidies up the library in many ways. It also improves how batch normalization is handled. Please check carefully the changes to see if any of this might impact your workflow. In particular, check out the vl_simplenn_tidy function to bring old models up to date. We also have opened a new discussion group for MatConvNet users (please use this group for discussions and GitHub for reporting bugs or similar). cuDNN v4 is also supported (although not all v4 features are used yet). Several pre-trained ImageNet models trained with MatConvNet using the supplied example code are now available.

New: 1.0-beta16 adds VGG-Face as a pretrained model.

New: Fully-Convolutional Networks (FCN) training and evaluation code is available here.

Obtaining MatConvNet

Documentation

Getting started

Use cases

Other information