Installation

The keras-texture package is hosted on GitHub. You can clone the repo with:

$ git clone https://github.com/rgmyr/keras-texture.git

The package can then be installed with the usual pip mechanisms:

$ cd keras-texture
$ pip install [-e] .

Requirements

  • keras (version >=2.0) with tensorflow backend
  • numpy
  • scikit-learn
  • scikit-image

The _cyvlfeat wrappers are required for use the texture.fisher module. They should be installed with conda, if at all possible.

TensorFlow is not explicitly required, due to the ambiguity between the tensorflow and tensorflow-gpu packages. Either is technically acceptable as a keras.backend.

Most functionality in the package will require a GPU, but some models (e.g., training an FV-CNN classifier on top of ImageNet pretrained models), can probably be used on a CPU-only machine.