AUTHOR=Zaccone Riccardo , Berton Gabriele , Masone Carlo 

TITLE=Distributed training of CosPlace for large-scale visual place recognition

JOURNAL=Frontiers in Robotics and AI

VOLUME=Volume 11 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2024.1386464

DOI=10.3389/frobt.2024.1386464

ISSN=2296-9144

ABSTRACT=Visual Place Recognition (VPR) is a popular computer vision task that aims to recognize the geographic location of a visual query, usually within a few meters of tolerance. Modern approaches address it from an image retrieval standpoint, using a kNN on top of embeddings extracted by a deep neural network from both the query and the images in a database. While most approaches rely on contrastive learning which limits their ability to train on large-scale datasets (due to mining), the recent work of CosPlace has introduced an alternative training paradigm using a classification task as proxy. This has been shown effective in expanding the potential of VPR models to learn from large-scale and fine-grained datasets. In this work we experimentally analyze CosPlace from a continual learning perspective and show that its sequential training procedure leads to sub-optimal results. As a solution, we propose a different formulation that not only effectively solves the pitfalls of the original training strategy, but also enables faster and more efficient distributed training. Finally, we discuss open challenges in further speeding up large-scale image retrieval for VPR.