Reproduce “Drag Your Gaussian: Effective Drag-Based Editing with Score Distillation for 3D 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
Yansong Qu, Dian Chen, Xinyang Li, Xiaofan Li, Shengchuan Zhang, Liujuan Cao, Rongrong Ji
Drag Your Gaussian (DYG) introduces an interactive framework for precise drag-based geometric editing on 3D Gaussian Splatting (3DGS) scenes. The method successfully performs complex deformations and transformations while maintaining high visual quality and multi-view consistency, significantly outperforming existing text-guided and naive 2D-lifting approaches in user studies and quantitative evaluations.
Key insights
- Combining implicit triplane representations with explicit 3D Gaussians effectively addresses the challenge of sparse Gaussians, enabling robust and continuous geometric deformation.
- A two-stage dragging process, separating geometric scaffold optimization from texture refinement, provides stability and prevents unintended global changes while ensuring local editing quality.
- Leveraging different specialized U-Nets within the Drag-SDS loss (inpainting for masked regions and a LoRA-trained U-Net for overall scene distribution) is critical for achieving multi-view consistent and artifact-free 3D edits from 2D LDM guidance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.