SlotLifter: Slot-guided Feature Lifting for Learning Object-centric Radiance Fields

本文提出了一种新颖的物体中心辐射模型SlotLifter，通过基于槽口的特征提升，同时解决场景重建和分解问题。这种设计将物体中心学习表示和基于图像的渲染方法结合起来，为四个具有挑战性的合成和四个复杂的真实世界数据集提供了最先进的场景分解和新视角合成性能，远远超过现有的3D物体中心学习方法。通过广泛的实验研究，我们展示了SlotLifter设计的有效性，并揭示了未来潜在方向的关键见解。

SlotLifter: Object-Centric Radiance Field Modeling and Rendering

SlotLifter is a novel object-centric radiance model that jointly addresses scene reconstruction and decomposition via slot-guided feature lifting, offering state-of-the-art performance in scene decomposition and novel-view synthesis on challenging synthetic and complex real-world datasets.

The paper proposes a new object-centric radiance model that combines object-centric learning representations and image-based rendering methods. The model outperforms existing 3D object-centric learning methods on both synthetic and real-world datasets. The paper includes extensive ablative studies that reveal key insights for potential future directions. The experiments are well-designed and the datasets used are challenging. The paper also provides open-source code for reproducibility.

Related work includes 'NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis' and 'GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis'.