kerfex

Generic feature extraction using keras pre-built CNN's with imagenet weights.

Getting Started

Dependencies

You need Python 3.7 or later to run kerfex. You can find it at python.org.

You aso need pandas, numpy, keras and tensorflow packages, which is available from PyPI. If you have PyPI, run:

pip install pandas numpy keras tf-nightly

Installation

Clone this repo to your local machine using:

git clone https://github.com/caiocarneloz/kerfex.git

Or install it using pip:

pip install kerfex

Usage

The demo.py file shows a simple example using VGG16 with three Unsplash images from the authors @mybibimbaplife, @davidbraud, and @analoglugunler. The "extract" function requires:

• CNN instance itself
• CNN pre-processing module
• List of images
• Images shape

As return, the function will send a pandas Dataframe containing the numerical features extracted from every image, where each line represents a single image and each column represents a single feature:

                0          1          2   ...      15599     15600     15601
	
0  	 0.000000   0.000000   0.000000   ...   0.000000  0.000000  3.754401
1  	 0.000000  15.284859  37.369953   ...  22.756908  6.398854  0.000000
2  	12.172541   0.000000   0.000000   ...   0.000000  0.000000  0.000000