Introduction

IVFADC implements an inverted file system with asymmetric distance computation for fast approximate nearest neighbor search in large i.e. billion-scale, high dimensional datasets.

Implemented features

• indexing: support for naive (brute search) and HNSW coarse quantizers
• adding points to the index: push!, pushfirst!
• removing points from the index: pop!, popfirst!, delete_from_index!
• searching into the index: knn_search
• saving/loading the index to/from disk: save_ivfadc_index, load_ivfadc_index

Installation

Installation can be performed from either inside or outside Julia.

Git cloning

$ git clone https://github.com/zgornel/IVFADC.jl

Julia REPL

The package can be installed from inside Julia with:

using Pkg
Pkg.add("IVFADC")

or

Pkg.add(PackageSpec(url="https://github.com/zgornel/IVFADC.jl", rev="master"))

for the latest master branch.

References

• Jègou et al. "Product quantization for nearest neighbor search", IEEE TPAMI, 2011
• Baranchuk et al. "Revisiting the inverted indices for billion-scale approximate nearest neighbors, ECCV, 2018"