Spherical Hashing

A Tensorflow 2 implementation of Spherical Hashing.

Spherical hashing is a way to compute a binary encoding of a feature vector while still maintaining spatial coherence. This binary encoding can then be used for an approximate nearest neighbor solution since the compactness of this representation allows for faster neighbor search.

Usage

from spherical_hashing import train_spherical_hashing
import tensorflow as tf

x_train = tf.random.uniform(shape=(1000, 512), minval=-10.0, maxval=10.0)
sph_model = train_spherical_hashing_model(x_train, n_bits=32)

x_test = tf.random.uniform(shape=(100, 512), minval=-10.0, maxval=10.0)
bits = sph_model(x_test, apply_pack_bits=True)

Installation

pip install tf-spherical-hashing

Development

Run tests

./scripts/run-tests.sh

Publish pip package

./scripts/publish.sh