Skip to content

Latest commit

 

History

History
112 lines (73 loc) · 3.7 KB

README.md

File metadata and controls

112 lines (73 loc) · 3.7 KB
Raw

Keras Model Zoo

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

Install

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

git clone https://github.com/albertomontesg/keras-model-zoo.git
cd keras-model-zoo
python setup.py install

Also as a pyp package:

pip install kerasmodelzoo

Usage

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
model.fit(X, Y)

Models Available

At this moment the models available are:

VGG

Reference:

@article{DBLP:journals/corr/SimonyanZ14a,
  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       = {http://arxiv.org/abs/1409.1556},
  timestamp = {Wed, 01 Oct 2014 15:00:05 +0200},
  biburl    = {http://dblp.uni-trier.de/rec/bib/journals/corr/SimonyanZ14a},
  bibsource = {dblp computer science bibliography, http://dblp.org}
}

Framework used: Caffe

License: unrestricted use

Dataset used to train: ILSVRC-2014

Description:

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.

C3D

Reference:

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

Description:

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.

Contribute

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