ScenePainter: Semantically Consistent Perpetual 3D Scene Generation with Concept Relation Alignment
Chong Xia, Shengjun Zhang, Fangfu Liu, Chang Liu, Khodchaphun Hirunyaratsameewong, Yueqi Duan
2025-07-29
Summary
This paper talks about ScenePainter, a method to create 3D scenes that look and feel consistent as they grow and change over time. It uses a special structure to understand how different parts of a scene are related so new parts can be added without losing the original meaning.
What's the problem?
The problem is called semantic drift, which means when you keep adding new views to a 3D scene, the details can slowly change or get mixed up, making the scene less accurate and less believable over time.
What's the solution?
ScenePainter fixes this by using a hierarchical graph called SceneConceptGraph that captures the relationships between objects and parts in the scene. This graph helps guide the process of adding new views in a way that keeps everything consistent. It also allows the scene to change and grow smoothly while keeping its original meaning intact.
Why it matters?
This matters because it helps create long, believable 3D scenes that can be used in video games, movies, virtual reality, and other areas where realistic, detailed, and stable 3D worlds are needed.
Abstract
ScenePainter addresses semantic drift in 3D scene generation by aligning scene-specific priors with current scene comprehension using a hierarchical graph structure.