Description

The package implements Improved Fisher Vectors as described in [1]. For a more concise description of Fisher Vectors see [2]. The functionality includes:

• Fitting a Gaussian Mixture Model (GMM)
• Determining the number of GMM components via BIC
• Saving and loading the fitted GMM
• Computing the (Improved) Fisher Vectors based on the fitted GMM

Installation

For intsallation via pip run the following command on your terminal (requires python 3.4 or higher):

$ pip install fishervector

afterwards you should be able to import the package in python:

from fishervector import FisherVectorGMM

First Steps

1. Simulate some data / get your data ready

We randomly sample data just for this tutorial -> use your own data e.g. SIFT features of images

import numpy as np
shape = [300, 20, 32] # e.g. SIFT image features
image_data = np.concatenate([np.random.normal(-np.ones(30), size=shape), np.random.normal(np.ones(30), size=shape)], axis=0)

2. Train the GMM

from fishervector import FisherVectorGMM
fv_gmm = FisherVectorGMM(n_kernels=10).fit(image_data)

Or alternatively fit the GMM using the BIC to determine the number of GMM components automatically:

from fishervector import FisherVectorGMM
fv_gmm = FisherVectorGMM().fit_by_bic(test_data, choices_n_kernels=[2,5,10,20])

3. Computing improved fisher vectors

image_data_test = image_data[:20] # use a fraction of the data to compute the fisher vectors
fv = fv_gmm.predict(image_data_test)

Contributors:

• Jonas Rothfuss (https://github.com/jonasrothfuss/)
• Fabio Ferreira (https://github.com/ferreirafabio/)

References:

• [1] https://www.robots.ox.ac.uk/~vgg/rg/papers/peronnin_etal_ECCV10.pdf
• [2] http://www.vlfeat.org/api/fisher-fundamentals.html