SAGE-VPR

Official checkpoints for SAGE: Spatial-visual Adaptive Graph Exploration for Efficient Visual Place Recognition.

Paper: https://arxiv.org/abs/2509.25723
GitHub: https://github.com/chenshunpeng/SAGE
Project Page: https://chenshunpeng.github.io/projects/SAGE
Demo: https://huggingface.co/spaces/shunpeng/sage-visual-place-recognition

Checkpoints

File	Description
SAGE.pth	Full SAGE checkpoint with the cross-image encoder.
SAGE_No-Encoder_Vit-B.pth	SAGE checkpoint without the cross-image encoder, using DINOv2 ViT-B as the backbone.
SAGE_No-Encoder_Vit-L.pth	SAGE checkpoint without the cross-image encoder, using DINOv2 ViT-L as the backbone.

Online Demo

An interactive demo is available on Hugging Face Spaces:

Try the SAGE Visual Place Recognition Demo

Acknowledgements

This interactive demo was originally created by @apolinario and the Hugging Face team.

We sincerely appreciate their support in making SAGE more accessible to the community.