SceneMaker: Open-set 3D Scene Generation with Decoupled De-occlusion and Pose Estimation Model
Reading time: 2 minute
...
📝 Original Info
- Title: SceneMaker: Open-set 3D Scene Generation with Decoupled De-occlusion and Pose Estimation Model
- ArXiv ID: 2512.10957
- Date: 2025-12-11
- Authors: Yukai Shi, Weiyu Li, Zihao Wang, Hongyang Li, Xingyu Chen, Ping Tan, Lei Zhang
📝 Abstract
Scene Images Different Views Generated by SceneMaker Open-set Scenes Indoor Scenes Figure 1. Our method not only achieves superior performance in both indoor and open-set scenarios but also demonstrates stronger generalization across synthetic and real-world captured images.📄 Full Content
Open-set 3D scene generation aims to synthesize 3D scenes containing arbitrary objects in any open-world domain from a single image. It is a fundamental task with high demand in AIGC and embodied AI, including applications such as 3D asset creation, simulation environment construction, and 3D perception for decision-making. However, limited scene datasets [4,13,18] have confined most existing methods [7,11,12,14,19,31,34,45] to constrained domains like indoor scenes.
Recently, the advent of large-scale 3D object datasets [15] has driven rapid progress in open-set 3D object generation models [26-28, 52, 54, 58, 60], and emerging methods [3,22,29,33,55] are beginning to extend scene generation toward open-set settings. Despite all the progress, existing methods still struggle to simultaneously produce high-quality geometry and accurate poses under severe occlusion and open-set settings in Figure 10.
…(본문이 길어 일부가 생략되었습니다.)
Reference
This content is AI-processed based on open access ArXiv data.