SceneMaker: Open-set 3D Scene Generation with Decoupled De-occlusion and Pose Estimation Model

Authors

Yukai Shi; Weiyu Li; Zihao Wang; Hongyang Li; Xingyu Chen; Ping Tan; Lei Zhang

Scores

Rationale

The paper introduces a novel decoupled framework for 3D scene generation that addresses the challenges of de-occlusion and pose estimation in open-set environments. This approach is technically significant as it seeks to improve both geometric quality and pose accuracy, which are critical in applications like AR/VR and robotics. Its potential transferability to other domains is moderate, as the techniques could be applicable to other 3D tasks, but the focus remains on scene generation. The work aligns well with current momentum in AI research on improving 3D scene understanding and generation. The evidence presented is robust, with comprehensive experiments and released datasets, supporting the claims. The approach has a reasonable chance of influencing future work in open-set 3D scene processing.