Reproduce “3DitScene: Editing Any Scene via Language-guided Disentangled Gaussian Splatting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qihang Zhang, Yinghao Xu, Chaoyang Wang, Hsin-Ying Lee, Gordon Wetzstein, Bolei Zhou, Ceyuan Yang
3DitScene introduces a framework for language-guided 3D scene editing from a single input image, enabling object-level manipulation with 3D consistency. The system combines disentangled 3D Gaussian Splatting with diffusion models and language models, demonstrating the ability to edit object appearance and layout while preserving scene coherence.
Key insights
- Combining 3D Gaussian Splatting with language models and diffusion models enables robust 3D scene representation and editing from a single image.
- Disentangling 3D objects through semantic feature distillation and layout augmentation allows for precise, language-guided object-level manipulation.
- Expanding the 3DGS representation with depth-aware inpainting helps generate consistent novel views, even for previously unobserved regions of the scene.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.