Auto-Regressive Diffusion Model Improves 3D Reconstruction
A new AI method called ArtiFixer uses auto-regressive diffusion models to enhance 3D reconstructions from 3D Gaussian Splatting. It addresses two key issues: scalability and quality. Existing methods use image diffusion or bidirectional video models limited in views per pass, requiring costly iterative distillation. ArtiFixer employs a two-stage pipeline with a novel opacity mixing strategy to encourage consistency, improving novel view synthesis in under-observed areas.
Key facts
- ArtiFixer is a two-stage pipeline using auto-regressive diffusion models.
- It enhances 3D Gaussian Splatting reconstructions.
- Addresses scalability and quality shortcomings of prior methods.
- Novel opacity mixing strategy encourages consistency.
- Improves novel view synthesis in under-observed areas.
- Published on arXiv with ID 2603.00492.
Entities
Institutions
- arXiv