Skip to content
Keras Model Zoo
Python Shell
Branch: master
Clone or download

Latest commit

albertomontesg Merge pull request #8 from albertomontesg/feature/pyp
Move the repository into a pyp package
Latest commit 249910d Jun 10, 2016


Type Name Latest commit message Commit time
Failed to load latest commit information.
.github Update Contributing Jun 10, 2016
kerasmodelzoo Remove old files and update README Jun 10, 2016
scripts/download Add new script for download vgg19 Jun 10, 2016
.gitignore Initial commit Apr 4, 2016 Remove old files and update README Jun 10, 2016
setup.cfg Merge pull request #8 from albertomontesg/feature/pyp Jun 11, 2016 Merge pull request #8 from albertomontesg/feature/pyp Jun 11, 2016

Keras Model Zoo

Repository to share all the models that the community has found and worked with the Keras framework. Official documentation here


To install this package you should first download this repository and then proceed with the installation:

git clone
cd keras-model-zoo
python install

Also as a pyp package:

pip install kerasmodelzoo


The usage is really easy. For each topology available you can load the model and also the mean which was trained with.

from kerasmodelzoo.models.vgg import vgg16

model = vgg16.model()
mean = vgg16.mean

It is also possible to load the weights or print the summary of the model if you give the parameters set to True:

from kerasmodelzoo.models.vgg import vgg16

model = vgg16.model(weights=True, summary=True)
mean = vgg16.mean
model.compile(loss='mse', optimizer='sgd')
X = X - mean, Y)

Models Available

At this moment the models available are:



  author    = {Karen Simonyan and
               Andrew Zisserman},
  title     = {Very Deep Convolutional Networks for Large-Scale Image Recognition},
  journal   = {CoRR},
  volume    = {abs/1409.1556},
  year      = {2014},
  url       = {},
  timestamp = {Wed, 01 Oct 2014 15:00:05 +0200},
  biburl    = {},
  bibsource = {dblp computer science bibliography,}

Framework used: Caffe

License: unrestricted use

Dataset used to train: ILSVRC-2014


This is the Keras model of the 16-layer network used by the VGG team in the ILSVRC-2014 competition. Project site. Gist where the model was obtained here.

It has been obtained by directly converting the Caffe model provived by the authors.

In the paper, the VGG-16 model is denoted as configuration D. It achieves 7.5% top-5 error on ILSVRC-2012-val, 7.4% top-5 error on ILSVRC-2012-test.

Please cite the paper if you use the models.



Tran, Du, et al. "Learning Spatiotemporal Features With 3D Convolutional Networks." Proceedings of the IEEE International Conference on Computer Vision. 2015.

Framework used: C3D (Caffe fork)

Dataset used to train: Sports1M


This model was trained using a modified version of BVLC Caffe to support 3-Dimensional Convolutional Networks. The C3D pre-trained model provided was trained on Sports-1M dataset and can be used to extract 3D-conv features.

Here are some results from the paper using the C3D features.

Dataset UCF101 ASLAN UMD-Scene YUPENN-Scene Object
C3D + linear SVM 82.3 78.3 (86.5) 87.7 98.1 22.3

If used this model, please refer to the citations on the project website.


On .github/ there is a detailed explanation about how to contribute to this repository with new models. Everyone is welcome and invited to participate.

You can’t perform that action at this time.