{"repo":"mingyu-hkustgz/RESQ","free":true,"listed":false,"github":"https://github.com/mingyu-hkustgz/RESQ","clone":"git clone https://github.com/mingyu-hkustgz/RESQ.git","description":"High-Ratio Vector Quantization","language":"C++","stars":15,"topics":["database","similarity-search","vector","vector-database"],"license":null,"category":"databases-storage","readme_excerpt":"Quantization Meets Projection: A Happy Marriage for Approximate k-Nearest Neighbor Search Introduction This is the official implementation of the early paper Fast High-dimensional Approximate Nearest Neighbor Search with Efficient Index Time and Space, and our VLDB 2026 version: Quantization Meets Projection: A Happy Marriage for Approximate k-Nearest Neighbor Search [Scalable Data Science]. MRQ leverages data distribution to achieve better distance correction and a higher vector compression ratio. It significantly outperforms state-of-the-art AKNN search methods based on graph or vector quantization, achieving up to a 3x efficiency speed-up with only 1/3 the length of quantized code while maintaining the same accuracy. Requirements C++17 Python OpenMP Directory Structure Reproduction Prepare Datasets All datasets we used for the evaluation can be downloaded from ANN-Benchmark or another public repo. Download the base and query set to the ./DATA directory and ensure they are in .fvecs / .ivecs format. Running Tests Step 1. Specify datasets in set.sh Step 2. Set indexing and searching parameters (Optional) To run tests on new datasets, you must set parameters for them. Indexing parameters: - RabitQ: - B : in ./test/RaBitQ/test index.sh , number of bits used in quantization. Determined by dimension D with B = (D + 63) // 64 64 - BB : in ./src/RaBitQ/index.cpp , same as B - DIM : in ./src/RaBitQ/index.cpp , dimension of the dataset. - MRQ/MRQ+: - B : in ./test/RESQ/test index.sh","default_branch":null,"files":null,"tree":[],"storefront":"/r/mingyu-hkustgz","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/mingyu-hkustgz/RESQ/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}