Generating Unbounded 3D Nature Scenes with Persistent Camera-Independent World Models (arxiv.org)

Summary:

• Recent 3D image generative models often operate on fixed-extent volumes with limited camera motions.
• A new approach synthesizes unbounded nature scenes, enabling large camera motion while maintaining a persistent 3D world model.

Key terms:

• 3D image generative models: Models that create 3D images with realistic image quality
• Unbounded nature scenes: Synthesizing limitless 3D landscapes for larger camera motions
• Persistent 3D world model: A consistent, camera-independent representation of a 3D scene
• Extendable, planar scene layout grid: A scene representation that can be rendered from any camera pose
• Panoramic skydome: A dome-like structure representing the sky in a 3D scene

Tags:

Research 3D image Unbounded Nature Scenes Camera Motion Scene Representation Planar Scene Layout Grid Single View Internet Photos Panoramic Skydome 3D landscapes persistent 3D world model