keras-image-helper 0.0.1

A lightweight library for pre-processing images for pre-trained keras models

Keras Image Helper

A lightweight library for pre-processing images for pre-trained keras models

Imagine you have a Keras model. To use it, you need to apply a certain pre-processing function to all the images. Something like that:

from tensorflow.keras.applications.xception import preprocess_input

What if you want to now deploy this model to AWS Lambda? Or deploy your model with TF-Serving? You don't want to use the entire TensorFlow package just for that.

The solution is simple - use keras_image_helper

Usage

For an xception model:

from keras_image_helper import create_preprocessor
preprocessor = create_preprocessor('xception', target_size=(299, 299))

url = 'http://bit.ly/mlbookcamp-pants'
X = preprocessor.from_url(url)

Now you can use X for your model:

preds = model.predict(X)

That's all :tada:

For more examples, check test.ipynb

Currently you can use the following pre-processors:

• xception
• resnet50
• vgg16
• inception_v3

If something you need is missing, PRs are welcome

Installation

It's available on PyPI, so you can install it with pip:

pip install keras_image_helper

Or with Pipenv:

pipenv install keras_image_helper

You can also install the latest version from this repo:

git clone [e-mail]:alexeygrigorev/keras-image-helper.git
python setup.py install

Publishing

Use twine for that:

pip install twine

Generate a wheel:

python setup.py sdist bdist_wheel

Check the packages:

twine check dist/*

Upload the library to test PyPI to verify everything is working:

twine upload --repository-url https://test.pypi.org/legacy/ dist/*