featurescope 0.0.1

FeatureScope Python Library.

FeatureScope (Python package)

This package provides a function apply_featurizer to compute custom features on images contained in a folder, so that these images and features can be visualized with the featurescope.

Example

We have a local folder of images:

/mnist
   01.png
   02.png
   ...

and a featurizer function in Python:

def mnist_featurizer(image: np.ndarray) -> Dict:
    return {
        "mean": np.mean(image),
        "max": np.max(image),
        "min": np.min(image),
    }

then, we can use:

from featurescope import apply_featurizer

apply_featurizer(mnist_featurizer, images_dir="./mnist")

Running apply_featurizer will load, and apply the featurizer function to all images in the folder, and save a CSV file dataset.csv in that same folder.

/mnist
   01.png
   02.png
   ...
   dataset.csv  <- Contains the computed image features!

With the features computed and stored, the folder /mnist can be dropped in the featurescope app in a web browser for visualization.

Installation

From PyPi:

pip install featurescope