Reproduce “ArtisanGS: Interactive Tools for Gaussian Splat Selection with AI and Human in the Loop” 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
Clement Fuji Tsang, Anita Hu, Or Perel, Carsten Kolve, Maria Shugrina
ArtisanGS, developed by NVIDIA and University of Toronto, provides interactive tools for flexibly segmenting objects within unstructured 3D Gaussian Splatting (3DGS) scenes, combining AI automation with user-in-the-loop correction. The system achieves competitive segmentation accuracy on the NVOS dataset while operating interactively within 1-5 seconds, enabling precise control over complex "in-the-wild" captures.
Key insights
- Combining AI-driven automation with extensive, interactive user control is crucial for practical 3DGS segmentation, bridging the gap between fully automatic and manual approaches.
- A fast, iterative workflow, enabled by processing times of 1-5 seconds, is essential for allowing users to efficiently diagnose and correct segmentation errors.
- Leveraging a differentiable 3DGS renderer as a "black box" for 3D mask aggregation allows for flexible and widely applicable segmentation without scene-specific pre-training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.