🏆 Winning NeurIPS' Competition Track: Big ANN, Practical Vector Search Challenge. (OOD Track) (Our other solution won the Sparse Track)

The fastest open-source cross-modal vector search solution.

MysteryANN is a cutting-edge research project centered around cross-modal retrieval. It has been specifically designed for cross-modal Approximate Nearest Neighbor Search (ANNS) while maintaining robustness for single-modal vector search. This repository builds upon the NSG repo and incorporates other open-source implementations. We have added our custom index codes extensively, resulting in significant speed improvements and state-of-the-art performance.

+ 274 (total downloads)

Please Note:

• This codebase is only intended for use in the NIPS 2023 Big-ANN Competition.

Our research paper is submitted, so we kindly say that this repository or idea does not support academic purposes temporarily. We appreciate your understanding. I plan to provide a tidy library in the near future.

Repository Organization:

• The codes for the NIPS 2023 challenge are available in separate branches.
• As we are currently focused on our paper, the codebase might not be as tidy as we'd like. Rest assured, we plan to clean up the code in the near future.

About the Name:

• The name "MysteryANN" is temporary and may not be the final name for this project; it's just for fun and represents "unknown."
• RoarGraph will be the suitable name.

(Unfinished) To get started with MysteryANN, please refer to the following resources:

• Installation Guide
• Usage Documentation

Un-decided

If you wish, please leave a message if you plan to use this idea.

For questions or inquiries, feel free to reach out to Meng Chen at [email protected] [email protected]