Scene Decomposition

Scene decomposition aims to separate complex scenes into their constituent parts, such as individual objects or semantic regions, enabling easier analysis and manipulation. Current research focuses on developing methods that leverage neural radiance fields (NeRFs) and other deep learning architectures, often incorporating self-supervised learning and multi-modal information (e.g., images and text) to achieve robust and accurate decomposition. This work is significant for advancing 3D scene understanding, enabling applications in areas like robotics, augmented reality, and computer-aided design through improved scene modeling, editing, and rendering.