Robust visual place recognition with adaptive deformable token aggregation

doi:10.1016/j.cag.2025.104342

Robust visual place recognition with adaptive deformable token aggregation

2025-08-06

0

PRE

AI

DOI：10.1016/j.cag.2025.104342

Original

Original request for help

Save

Abstract

En 中文

• Proposes ADTA-VPR, a robust framework for visual place recognition under complex conditions. • Introduces Deformable Token Merging to adaptively downsample features with spatial flexibility. • Designs Prototype-guided Token Aggregation to suppress redundancy and enhance distinctiveness. • Constructs FCBR, a VPR benchmark with realistic photographic degradations. • Achieves state-of-the-art performance on diverse and challenging VPR benchmarks.

Keywords:

ADTA-VPR

Deformable Token Merging

Prototype-guided Token Aggregation

FCBR benchmark

Visual Place Recognition

Robust visual place recognition with adaptive deformable token aggregation

Abstract

Journal

Organization