KhrulkovV/geometry-score

Geometry Score: A Method For Evaluating Generative Adversarial Networks

Python implementation of the algorithms from the paper. If you use this algorithm in your research we kindly ask you to cite our work

@article{khrulkov2018geometry,
  title={Geometry {S}core: {A} {M}ethod {F}or {C}omparing {G}enerative {A}dversarial {N}etworks},
  author={Khrulkov, Valentin and Oseledets, Ivan},
  journal={arXiv preprint arXiv:1802.02664},
  year={2018}
}

Prerequisites

• Python 2.7 or Python 3.3+
• SciPy
• NumPy
• matplotlib
• GUDHI
• Cython

Basic usage

import numpy as np
import gs
X = np.random.rand(1000, 2)
rlt = gs.rlts(X, L_0=32, gamma=1.0/8, i_max=100, n=100)
mrlt = np.mean(rlt, axis=0)

For more details see the MNIST example and toy examples .