| 1 |
|
| 2 |
|
| 3 |
README for the link & code implementation |
| 4 |
========================================= |
| 5 |
|
| 6 |
What is this? |
| 7 |
------------- |
| 8 |
|
| 9 |
Link & code is an indexing method that combines HNSW indexing with |
| 10 |
compression and exploits the neighborhood structure of the similarity |
| 11 |
graph to improve the reconstruction. It is described in |
| 12 |
|
| 13 |
``` |
| 14 |
@inproceedings{link_and_code, |
| 15 |
author = {Matthijs Douze and Alexandre Sablayrolles and Herv\'e J\'egou}, |
| 16 |
title = {Link and code: Fast indexing with graphs and compact regression codes}, |
| 17 |
booktitle = {CVPR}, |
| 18 |
year = {2018} |
| 19 |
} |
| 20 |
``` |
| 21 |
|
| 22 |
ArXiV [here](https://arxiv.org/abs/1804.09996) |
| 23 |
|
| 24 |
The necessary code for this paper was removed from Faiss in version 1.8.0. |
| 25 |
For a functioning version, use Faiss 1.7.4. |