Avila Herrera, Eduardo Andres (National Univesity Ireland Maynooth), McCarthy, Tim (National Univesity Ireland Maynooth), McDonald, John (Maynooth University)

Cross View Shared Transformer (CVST) a Shared Weight Siamese Network for Limited FoV Image Based Localisation

Scheduled for presentation during the Poster Session "Transformer networks" (WeBT13), Wednesday, September 25, 2024, 14:30−16:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Cross view localisation (CVL) has received significant attention in recent years due to its potential applications in robotics and autonomous driving. However, research in the area has relied heavily on panoramic images to improve recall performance. In this paper, we focus on the CVL task where ground level images are limited to a 90º field-of-view (FoV). Extensive testing on multiple network architectures is performed to quantify the impact of the loss function and the effect of weight sharing in Siamese Networks for CVL tasks. Additionally, we introduce the Cross View Shared Transformer (CVST) net work, a novel Siamese Network architecture employing shared weights capable of extracting features from different view points without the aid of external projection. CVST demonstrates state of the art performance on KITTI a real-world image retrieval benchmark using limited field of view imagery, while reducing the complexity of the Siamese Network.