BetterScene: 3D Scene Synthesis with Representation-Aligned Generative Model

BetterScene is a new AI approach that enhances 3D scene synthesis and novel view generation from sparse photos by leveraging Stable Video Diffusion with improved regularization techniques. The method integrates 3D Gaussian Splatting and addresses consistency issues in existing diffusion-based solutions through temporal equivariance and vision foundation model alignment.

• →BetterScene improves novel view synthesis quality using extremely sparse, unconstrained photos as input.
• →The approach builds on Stable Video Diffusion model pretrained on billions of frames for enhanced backbone performance.
• →Two key innovations include temporal equivariance regularization and vision foundation model-aligned representation applied to the VAE module.
• →The method integrates 3D Gaussian Splatting to generate continuous, artifact-free novel views.
• →Testing on DL3DV-10K dataset shows superior performance compared to existing state-of-the-art methods.