TSCM: A Teacher-Student Model for Vision Place Recognition Using Cross-Metric Knowledge Distillation

Visual place recognition (VPR) plays a pivotal role in autonomous explorationand navigation of mobile robots within complex outdoor environments. Whilecost-effective and easily deployed, camera sensors are sensitive to lightingand weather changes, and even slight image alterations can greatly affect VPRefficiency and precision. Existing methods overcome this by exploiting powerfulyet large networks, leading to significant consumption of computationalresources. In this paper, we propose a high-performance teacher and lightweightstudent distillation framework called TSCM. It exploits our devisedcross-metric knowledge distillation to narrow the performance gap between theteacher and student models, maintaining superior performance while enablingminimal computational load during deployment. We conduct comprehensiveevaluations on large-scale datasets, namely Pittsburgh30k and Pittsburgh250k.Experimental results demonstrate the superiority of our method over baselinemodels in terms of recognition accuracy and model parameter efficiency.Moreover, our ablation studies show that the proposed knowledge distillationtechnique surpasses other counterparts. The code of our method has beenreleased at https://github.com/nubot-nudt/TSCM.