This is diffusion for retrieval instead of image generation. The slides summarize some famous papers, and the code is an easy and robust implementation.
This tutorial is mainly based on Iscen's CVPR17 paper) and the code of Dmytro Mishkin.
I find this version gives better performance than https://github.com/fyang93/diffusion, which is also a very good contribution.
The two most famous reranking methods, diffusion and spatial verification, can be combined.
Check: https://github.com/anguoyuan/Hypergraph-Propagation-and-Community-Selection-for-Objects-Retrieval (The best open-code mAP on ROxford and RParis. Fast)