chuanxiao1983/HVS

HVS: Hierarchical Graph Structure Based on Voronoi Diagrams for Solving Approximate Nearest Neighbor Search

Project Title

HVS: Hierarchical Graph Structure Based on Voronoi Diagrams for Solving Approximate Nearest Neighbor Search

Getting Started

Prerequisites

• OpenCV 3.30+
• GCC 4.9+ with OpenMP
• CMake 2.8+
• Boost 1.55+
• TCMalloc

Datasets and query sets

• Tiny80M (https://drive.google.com/file/d/1PzW9cqi8VbzH9Bu_7UDWAXjA2DlBorG3/view?usp=sharing)
• Other real datasets (https://www.cse.cuhk.edu.hk/systems/hash/gqr/datasets.html)

Compile On Ubuntu

Complie HVS (based on NSW)

$ cd hnsw/
$ mkdir build/ && cd build/
cmake ..
make 

Commands

• K is the value of top-K, L is the value of efsearch and qn is the size of query set
• T and delta are user-specified parameters of HVS
• The data set, query set and the ground_truth set are stored in ${dPath}.ds, ${qPath}.q and truth.gt

Build HVS index

./hnsw/build/main ${dPath}.ds nullptr ${n} ${d} ${T} -1 ${delta} -1

Search in HVS

./hnsw/build/main ${dPath}.ds ${qPath}.q ${n} ${d} ${T} ${qn} ${K} ${L}

A running example (ImageNet)

• Donwload the dataset, query set and the ground truth set of ImageNet from the following link https://drive.google.com/file/d/1WV78sZT1j1oz-GoZTQdyKaNQgulRLe2O/view?usp=sharing
• Put all three files in the main folder
• Run the script

bash run_image.sh