Collaborative Visual Place Recognition Through Federated Learning

Visual Place Recognition (VPR) aims to estimate the location of an image bytreating it as a retrieval problem. VPR uses a database of geo-tagged imagesand leverages deep neural networks to extract a global representation, calleddescriptor, from each image. While the training data for VPR models oftenoriginates from diverse, geographically scattered sources (geo-tagged images),the training process itself is typically assumed to be centralized. Thisresearch revisits the task of VPR through the lens of Federated Learning (FL),addressing several key challenges associated with this adaptation. VPR datainherently lacks well-defined classes, and models are typically trained usingcontrastive learning, which necessitates a data mining step on a centralizeddatabase. Additionally, client devices in federated systems can be highlyheterogeneous in terms of their processing capabilities. The proposed FedVPRframework not only presents a novel approach for VPR but also introduces a new,challenging, and realistic task for FL research, paving the way to other imageretrieval tasks in FL.