Close, But Not There: Boosting Geographic Distance Sensitivity in Visual Place Recognition

Sergio Izquierdo, Javier Civera

Code and models for the ECCV 2024 paper "Close, But Not There: Boosting Geographic Distance Sensitivity in Visual Place Recognition" (CliqueMining)

Summary

In this repo, we include a novel mining pipeline, CliqueMining, that creates very difficult batches. It creates a graph of very similar images and samples cliques (representing places) to create challenging batches. This technique improves performance on many common datasets.

For more details, check the paper.

Weights

You can download the weights of the trained model here. To evaluate, follow the same steps as with SALAD.