.png){kind=icon}
.png){kind=icon}
Return
Web3D scene construction assisted by semantic segmentation
.png){kind=icon}
OA
AI
DOI：10.1080/10095020.2026.2667713.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
With the continuous evolution of network infrastructure, desktop applications are increasingly being migrated to web-based platforms. This transformation brings significant advantages, including seamless cross-platform accessibility and the elimination of plug-in dependencies, thereby enabling intuitive and efficient access to the spatial layout and current conditions of Web3D environments. Despite these benefits, limitations in browser computing resources pose notable challenges in achieving a balance between visual realism and scene dynamism in current Web3D construction methods. To overcome these constraints, this study introduces a novel approach that integrates semantic segmentation with procedural content generation (PCG) to facilitate rapid Web3D scene construction. Specifically, we adopt the DeepLabV3+ framework to perform semantic segmentation on two distinct datasets, incorporating multiple backbone networks and embedding an attention mechanism to enhance the delineation of fine-grained features, particularly at object boundaries. Following segmentation, PCG techniques are utilized to generate and visualize key environmental elements, including terrain, water bodies, and vegetation. These elements are further enriched by integrating semantic labels with corresponding 3D models and attribute data. Experimental validations demonstrate that the proposed method enables efficient and realistic Web3D scene creation while preserving dynamic behavior, effectively addressing diverse application scenarios. The results highlight the method’s potential as a practical and scalable solution for next-generation Web3D scene development.
Keywords:
Semantic segmentation
three-dimensional scene construction
visualization
web-based three-dimensional scenes
Journal
G
IF:
5.5
Papers:
838
Citations:
2.4K