Citation

Ching, Jiun Yen and Wong, Lai Kuan and Kung, Fabian Wai Lee (2025) Semantic Scene Completion from a Single Depth Image with Coarse-Grained Segmentation. In: 2025 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), 22-24 October 2025, Singapore, Singapore.

Text 472.pdf - Published Version Restricted to Repository staff only Download (2MB)

Abstract

Semantic Scene Completion (SSC) plays an important role in computer vision applications such as mobile robots navigation. SSC aims to reconstruct a complete 3D volumetric scene from an incomplete 3D scene in the form of a single depth image, by assigning each voxel in the camera frustum a class label. The main challenge lies in directly predicting the fine-grained classes in a complex and class-diverse indoor environment. To resolve this challenge, we first introduce a novel class taxonomy for SUNCG scene labels based on common shared characteristics, whereby the hierarchy in the taxonomy describes the relationship between coarse-grained and fine-grained classes. We then designed a lightweight Coarse-Grained Segmentation (CGS) module, an adaptable decoder, to exploit the structured class relations and explicitly model the high- and mid-level features. We implemented and validated the CGS module with three existing SSC models, i.e., SSCNet, EdgeNet-D and Real-Time. Experimental results demonstrate that the incorporation of the CGS module improved the performance of fine-grained semantic segmentation of existing SSC models on NYUv2 and NYUCAD datasets, demonstrating the effectiveness of our approach.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Adaptation models, Three-dimensional displays, Navigation, Semantic segmentation, Taxonomy, Semantics, Robot vision systems, Real-time systems, Decoding, Mobile robots
Subjects:	T Technology > TR Photography
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	19 Mar 2026 01:00
Last Modified:	19 Mar 2026 01:00
URII:	http://shdl.mmu.edu.my/id/eprint/15497

Downloads

Edit (login required)