This paper presents a method to more efficiently group the images from a large-scale visual place recognition dataset. To utilize the large-scale dataset, previous studies employ GPS labels to quan- tize places into several groups, which is referred to as grouping, and cast them into classification problems. Our method addresses a limitation ob- served in the previous strategy, where the focus was solely on mitigating intra-group quantization errors, not considering the inter-group effects. We apply distance-based grouping to reduce redundant images between groups and by reducing the number of redundant images, we not only save memory and storage but also reduce the time for convergence. In this paper, we examine our grouping strategy in various test datasets and find out the best grouping strategy for large-scale visual place recognition datasets.
Efficient Data Grouping for Large-Scale Visual Place Recognition Dataset
Lect. Notes in Networks, Syst.
International Conference on Robot Intelligence Technology and Applications ; 2023 ; Taicang December 06, 2023 - December 08, 2023
2024-11-22
12 pages
Article/Chapter (Book)
Electronic Resource
English
Visual grouping and object recognition
IEEE | 2001
|Visual Grouping and Object Recognition
British Library Conference Proceedings | 2001
|Trajectory-Based Place-Recognition for Efficient Large Scale Localization
British Library Online Contents | 2017
|CMM: LiDAR-Visual Fusion with Cross-Modality Module for Large-Scale Place Recognition
SAE Technical Papers | 2023
|