D$^{2}$-VPR: A Parameter-efficient Visual-foundation-model-based Visual Place Recognition Method via Knowledge Distillation and Deformable Aggregation

2025-11-16

AI Summary

Visual place recognition (VPR) faces significant challenges in deploying large foundation models—e.g., DINOv2—on resource-constrained edge devices due to their excessive parameter count and computational overhead. To address this, we propose an efficient knowledge distillation and deformable aggregation framework. First, we design a two-stage distillation strategy augmented with a Distillation Recovery Module (DRM) to enhance feature-space alignment between teacher and student models. Second, we introduce a top-down attention-driven deformable aggregator (TDDA) that dynamically focuses on salient structural regions for robust place matching. Our approach preserves DINOv2’s strong representational capacity while drastically reducing model size and complexity. Experiments demonstrate that our method reduces parameters by 64.2% and floating-point operations by 62.6% compared to the state-of-the-art CricaVPR, with negligible degradation in VPR accuracy—thereby markedly improving feasibility for edge deployment.

Technology Category

• Computer Vision: Other Foundations of Computer Vision
• Machine Learning: Learning on the Edge & Model Compression
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Visual Place Recognition (VPR) aims to determine the geographic location of a query image by retrieving its most visually similar counterpart from a geo-tagged reference database. Recently, the emergence of the powerful visual foundation model, DINOv2, trained in a self-supervised manner on massive datasets, has significantly improved VPR performance. This improvement stems from DINOv2's exceptional feature generalization capabilities but is often accompanied by increased model complexity and computational overhead that impede deployment on resource-constrained devices. To address this challenge, we propose $D^{2}$-VPR, a $D$istillation- and $D$eformable-based framework that retains the strong feature extraction capabilities of visual foundation models while significantly reducing model parameters and achieving a more favorable performance-efficiency trade-off. Specifically, first, we employ a two-stage training strategy that integrates knowledge distillation and fine-tuning. Additionally, we introduce a Distillation Recovery Module (DRM) to better align the feature spaces between the teacher and student models, thereby minimizing knowledge transfer losses to the greatest extent possible. Second, we design a Top-Down-attention-based Deformable Aggregator (TDDA) that leverages global semantic features to dynamically and adaptively adjust the Regions of Interest (ROI) used for aggregation, thereby improving adaptability to irregular structures. Extensive experiments demonstrate that our method achieves competitive performance compared to state-of-the-art approaches. Meanwhile, it reduces the parameter count by approximately 64.2% and FLOPs by about 62.6% (compared to CricaVPR).Code is available at https://github.com/tony19980810/D2VPR.

Problem

• Reducing model complexity of visual foundation models for efficient deployment
• Minimizing knowledge transfer loss between teacher and student models
• Improving adaptability to irregular structures through deformable aggregation

Innovation

• Knowledge distillation reduces model parameters
• Deformable aggregation adaptively adjusts regions of interest
• Two-stage training aligns teacher-student feature spaces