ROSA-Net: Rotation-Robust Structure-Aware Network for fine-grained 3D shape retrieval

Fu, Rao, Zhang, Yunchi, Yang, Jie, Sun, Jiawei, Zhang, Fang-Lue, Lai, Yu-Kun ORCID: https://orcid.org/0000-0002-2094-5680 and Gao, Lin 2024. ROSA-Net: Rotation-Robust Structure-Aware Network for fine-grained 3D shape retrieval. Presented at: Computational Visual Media Conference 2024, Wellington, New Zealand, 10-12 April 2024. Computational Visual Media. CVM 2024. Lecture Notes in Computer Science , vol.14592 Singapore: Springer, pp. 295-319. 10.1007/978-981-97-2095-8_16

Preview

PDF - Accepted Post-Print Version Download (3MB) | Preview

Abstract

Fine-grained 3D shape retrieval aims to retrieve 3D shapes similar to a query shape in a repository with models belonging to the same class, which requires shape descriptors to represent detailed geometric information to discriminate shapes with globally similar structures. Moreover, 3D objects can be placed with arbitrary positions, orientations, and scales in real-world applications, which further requires shape descriptors to be robust to rotation and sensitive to scale. The shape descriptions used in existing 3D shape retrieval systems fail to meet the above two criteria. In this paper, we introduce a novel deep architecture, ROSA-Net, which learns rotation-robust and scale-sensitive 3D shape descriptors capable of encoding fine-grained geometric information and structural information, and thus achieve accurate results on the task of fine-grained 3D object retrieval. ROSA-Net extracts a set of compact and detailed geometric features part-wisely and discriminatively estimates the contribution of each semantic part to shape representation. Furthermore, our method can learn the importance of geometric and structural information of all the parts when generating the final compact latent feature of a 3D shape for fine-grained retrieval. We also build and publish a new 3D shape dataset with sub-class labels for validating the performance of fine-grained 3D shape retrieval methods. Qualitative and quantitative experiments show that our ROSA-Net outperforms state-of-the-art methods on the fine-grained object retrieval task, demonstrating its capability in geometric detail extraction. The code is available in the supplementary material.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Published Online
Status:	Published
Schools:	Computer Science & Informatics
Publisher:	Springer
ISBN:	978-981-97-2094-1
Date of First Compliant Deposit:	5 June 2024
Date of Acceptance:	5 December 2023
Last Modified:	11 Jul 2024 01:30
URI:	https://orca.cardiff.ac.uk/id/eprint/169482

Actions (repository staff only)

Edit Item