EFANNA

Category: SDKs & Libraries
Tags: ann, high-performance, vector-indexing
Source: https://github.com/ZJULearning/efanna

Overview

EFANNA is a C++ library for extremely fast approximate nearest neighbor (ANN) search on large-scale data. It implements the EFANNA algorithm described in the paper “EFANNA: Extremely Fast Approximate Nearest Neighbor Search Algorithm Based on kNN Graph” and provides both ANN search and approximate kNN graph construction.

Features

• Approximate nearest neighbor search

  • High-performance ANN search on large-scale vector datasets.
  • Designed as a building block for custom vector search and retrieval infrastructure.

• kNN graph construction

  • Efficient construction of approximate k-nearest neighbor graphs.
  • Can be used as a base structure for various graph-based search algorithms.

• Flexible initialization structures

  • Supports multiple hierarchical structures for search initialization, including:
    • Randomized KD-trees.
    • Random projection trees.
    • Hierarchical clustering trees.
    • Multi-table hashing (via the provided hashing samples).

• Algorithm framework

  • Implements the EFANNA framework as described in the associated research paper.
  • Separated modules under algorithm, general, and samples directories for experimentation and extension.

• Language and ecosystem

  • C++ implementation (efanna.hpp, Makefiles provided).
  • Example code and usage patterns under samples/.
  • Additional MATLAB-related utilities in the matlab/ directory.

• Documentation and examples

  • docs/ folder for algorithm and usage documentation (as provided in the repository).
  • Sample programs demonstrating index construction and query workflows.

• Performance focus

  • Designed specifically for speed on ANN search and kNN graph construction tasks.
  • Benchmark references and figures for ANN search performance on standard datasets (e.g., SIFT) included in the repository.

Pricing

EFANNA is an open-source library hosted on GitHub. No pricing or paid plans are specified in the provided content.

License

• A LICENSE file is present in the repository.
• Consult the repository’s LICENSE file for the exact license terms and conditions.