OneWorld: Taming Scene Generation with 3D Unified Representation Autoencoder

Dimensions overall score 8.0

• OneWorld generates 3D scenes with superior cross-view consistency compared to state-of-the-art 2D-based methods.result 90%
  Evidencepartial

  Comprehensive experiments demonstrate that OneWorld generates high-quality 3D scenes with superior cross-view consistency compared to state-of-the-art 2D-based methods.

  Implicationpartial

  Directly stated in abstract with comprehensive experiments mentioned

  Verificationpartial

  partial

• Existing diffusion-based 3D scene generation methods primarily operate in 2D image/video latent spaces, making cross-view consistency inherently challenging.limitation 85%
  Evidencepartial

  Existing diffusion-based 3D scene generation methods primarily operate in 2D image/video latent spaces, which makes maintaining cross-view appearance and geometric consistency inherently challenging.

  Implicationpartial

  Directly stated in abstract as motivation for the research

  Verificationpartial

  partial

• The 3D Unified Representation Autoencoder (3D-URAE) injects appearance and distills semantics into a unified 3D latent space.method 88%
  Evidencepartial

  Central to our approach is the 3D Unified Representation Autoencoder (3D-URAE); it leverages pretrained 3D foundation models and augments their geometry-centric nature by injecting appearance and distilling semantics into a unified 3D latent space.

  Implicationpartial

  Directly described in abstract as core technical component

  Verificationpartial

  partial

• Token-level Cross-View-Correspondence (CVC) consistency loss explicitly enforces structural alignment across views.method 87%
  Evidencepartial

  Furthermore, we introduce token-level Cross-View-Correspondence (CVC) consistency loss to explicitly enforce structural alignment across views

  Implicationpartial

  Directly stated as a technical innovation in the abstract

  Verificationpartial

  partial

• Manifold-Drift Forcing (MDF) mitigates train-inference exposure bias and shapes a robust 3D manifold.method 86%
  Evidencepartial

  and propose Manifold-Drift Forcing (MDF) to mitigate train-inference exposure bias and shape a robust 3D manifold by mixing drifted and original representations.

  Implicationpartial

  Directly described as a technical innovation in the abstract

  Verificationpartial

  partial

• The framework performs diffusion directly within a coherent 3D representation space.technical 90%
  Evidencepartial

  To bridge this gap, we present OneWorld, a framework that performs diffusion directly within a coherent 3D representation space.

  Implicationpartial

  Directly stated as the core approach in the abstract

  Verificationpartial

  partial

• The method leverages pretrained 3D foundation models as part of its architecture.technical 89%
  Evidencepartial

  it leverages pretrained 3D foundation models and augments their geometry-centric nature

  Implicationpartial

  Explicitly stated in abstract as a key component

  Verificationpartial

  partial

• The computational cost of training and inference in 3D latent space could be high, limiting scalability for real-time applications.limitation 75%
  Evidencepartial

  The computational cost of training and inference in 3D latent space could be high, limiting scalability for real-time applications.

  Implicationpartial

  Implied in analysis caveats but not directly measured in paper

  Verificationpartial

  partial

Competitive landscape

OneWorld is a framework for generating high-quality 3D scenes with superior cross-view consistency using a unified representation autoencoder.