Hello,
similar to caffe, keras should have model zoo so that people can replicate results of others. does anyone agree on that?
There's an example folder, so replicating is quiet easy to follow: https://github.com/fchollet/keras/tree/master/examples
Similar to lasagna's recipe: https://github.com/Lasagne/Recipes
There is also work in progress to load caffe models into keras.
@alvations : model zoo is for trained models not for the recipes/scripts. not everyone has powerful GPUs and it would be good to have model files (like VGGs in caffe).
@EderSantana : it would be nice to have such thing. but what I meant is a platform to share trained models (h5 file and architecture jsons).
@wolet have your head about https://github.com/sklearn-theano/sklearn-theano they have some good models pre-trained and you don't need GPU to test them. Since they also use theano for backend, all that we need is a Keras wrapper so we can fine tune and test them with Keras API (Volunteers Needed!!!).
There is a VGG16 model for Keras.
@willard-yuan great! we should have more!
This is a duplicate of #703.
There is also work in progress to load caffe models into keras.
@EderSantana Is there an update?
Most helpful comment
There is also work in progress to load caffe models into keras.