Pretrained models

This section describes how pre-trained models can be downloaded and used in MatConvNet. Using the pre-trained model is easy; just start from the example code included in the quickstart guide.

Remark: The following CNN models may have been imported from other reference implementations and are equivalent to the originals up to numerical precision. However, note that:

  1. Images need to be pre-processed (resized and cropped) before being submitted to a CNN for evaluation. Even small differences in the preprocessing details can have a non-negligible effect on the results.

  2. The example below shows how to evaluate a CNN, but does not include data augmentation or encoding normalization as for example provided by the VGG code. While this is easy to implement, it is not done automatically here.

  3. These models are provided here for convenience, but please credit the original authors.

Object detection

These models are trained for object detection in PASCAL VOC.

The model performance is as follows (mAP 11 indicates mean average precision computed using 11 point interpolation, as per PASCAL VOC 07 specification):

model training set PASCAL07 test mAP mAP 11
fast-rcnn-caffenet-pascal07-dagnn imnet12+pas07 57.3 % 58.1 %
fast-rcnn-vggm12-pascal07-dagnn imnet12+pas07 59.4 % 60.5 %
fast-rcnn-vgg16-pascal07-dagnn imnet12+pas07 67.3 % 68.7 %

Face recognition

These models are trained for face classification and verification.

Semantic segmentation

These models are trained for semantic image segmentation using the PASCAL VOC category definitions.

ImageNet ILSVRC classification

These modes are trained to perform classification in the ImageNet ILSVRC challenge data.

This is a summary of the performance of these models on the ILSVRC 2012 validation data:

model introduced top-1 err. top-5 err. images/s
resnet-50-dag 2015 24.6 7.7 396.3
resnet-101-dag 2015 23.4 7.0 247.3
resnet-152-dag 2015 23.0 6.7 172.5
matconvnet-vgg-verydeep-16 2014 28.3 9.5 200.9
vgg-verydeep-19 2014 28.7 9.9 166.2
vgg-verydeep-16 2014 28.5 9.9 200.2
googlenet-dag 2014 34.2 12.9 770.6
matconvnet-vgg-s 2013 37.0 15.8 586.2
matconvnet-vgg-m 2013 36.9 15.5 1212.5
matconvnet-vgg-f 2013 41.4 19.1 2482.7
vgg-s 2013 36.7 15.3 560.1
vgg-m 2013 37.3 15.9 1025.1
vgg-f 2013 41.1 18.8 1118.9
vgg-m-128 2013 40.8 18.4 1031.3
vgg-m-1024 2013 37.8 16.1 958.5
vgg-m-2048 2013 37.1 15.8 984.2
matconvnet-alex 2012 41.8 19.2 2133.3
caffe-ref 2012 42.6 19.7 1071.7
caffe-alex 2012 42.6 19.6 1379.8

Important notes:

File checksums

The following table summarizes the MD5 checksums for the model files.

MD5 File name
9dcc29b03edb5b136fda31fcd59bd025 fast-rcnn-caffenet-pascal07-dagnn.mat
961f997d7ff922f1ff6b7b20ad677a4c fast-rcnn-vgg16-pascal07-dagnn.mat
f09a662aef88ac4c23d294eb1fb6f385 fast-rcnn-vggm1k-pascal07-dagnn.mat
b5152a54bf61100a8ed61827c76e7d08 imagenet-caffe-alex.mat
675dd47906d34f4d9a70fc44210d3804 imagenet-caffe-ref.mat
97f4e1aa9ad6ed33fd325835710c5092 imagenet-googlenet-dag.mat
0998b7f66bd8dcbb78a3e75aaa6127f3 imagenet-matconvnet-alex.mat
790e116ec817a58960b103514891fdbf imagenet-matconvnet-vgg-f.mat
7e07705fc60c178ee3d2c09dab33acf3 imagenet-matconvnet-vgg-m.mat
82d5705d292714d0d6581a834485705b imagenet-matconvnet-vgg-s.mat
89a117515f66c3d7b0eb9d0516c65141 imagenet-matconvnet-vgg-verydeep-16.mat
16ff04c36c7bd33510e4a81db6dc764b imagenet-resnet-101-dag.mat
f399bef82f5bcaf290d07ccc818833d7 imagenet-resnet-152-dag.mat
d72aa76a8ddf8bd96e405ac0ba827724 imagenet-resnet-50-dag.mat
3513562d28089bd965cc050cbf8597a6 imagenet-vgg-f.mat
1410d01878346f911e991dd0c1bb983b imagenet-vgg-m-1024.mat
3a8dc312a44a99d21ad43e8d96a8590f imagenet-vgg-m-128.mat
087ec812e0a09980bd934e2e7ba157f6 imagenet-vgg-m-2048.mat
29294d5f62578c96a3533859514235f2 imagenet-vgg-m.mat
dbe23da22e82078debda75842b16d5fa imagenet-vgg-s.mat
f72d927587ca4c97fbd165ec9cb9997f imagenet-vgg-verydeep-16.mat
106118b7cf60435e6d8e04f6a6dc3657 imagenet-vgg-verydeep-19.mat
2e49dd427829cdbc08a0154f994687d7 pascal-fcn16s-dag.mat
a1331885ec72a0721e51ac9d16262a48 pascal-fcn32s-dag.mat
0eeb6a4bc819616ea66f88a3db878983 pascal-fcn8s-dag.mat
b0f7bd5833b555c6241dd0c05897ca41 pascal-fcn8s-tvg-dag.mat
3d6cd504bf9c98af4a561aad059565d1 vgg-face.mat

Older file versions

Older models for MatConvNet beta16 are available here. They should be numerically equivalent, but in beta17 the format has changed slightly for SimpleNN models. Older models can also be updated using the vl_simplenn_tidy function.