The authors built GraphWrit3R, an end-to-end model that encodes point cloud inputs via Sonata and Gaussian Splat inputs via Chorus, projects both onto a shared voxel grid, fuses them using a per-voxel contrastive alignment loss, and decodes the fused representation with a large language model to generate the scene graph.
End-to-end model writes 3D scene graphs directly from point clouds or Gaussian splats
GraphWrit3R outputs a structured JSON script listing objects, attributes, and relationships, bypassing multi-stage pipelines and ground-truth annotations.
Big Tech
Luka Milivojevic · Nikola Popovic · Sayan Deb Sarkar · Sebastian Koch · Iro Armeni · Luc Van Gool · +1 more
INSAIT, Sofia University · Stanford University · Ulm University
Research Digest··2 min read
Milivojevic et al.
Why this paper
From INSAIT, Sofia University and 2 others
In one line
GraphWrit3R generates 3D scene graphs directly from point clouds or Gaussian splats using an end-to-end LLM-based model.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§