Research map / 3D vision / 3D generation
DepWorldSG: Depth-Aware 3D Semantic Scene Graph Generation via World-Model Priors
Seok-Young Kim, Abdelrahman Elskhawy, TAEWOOK HA, Dooyoung Kim, Eunjae Shin, Benjamin Busam, Woontack Woo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add box-conditioned instance masking for depth sampling (ablation) | 2D lifting baseline · 3DSSG | Relationship recall: 28.5 -> 40.3 | Improved |
| add box-conditioned instance masking for depth sampling (ablation) | 2D lifting baseline · 3DSSG | Object recall: 63.4 -> 71.7 | Improved |
| add dual-domain depth refinement (DR) (ablation) | DeWorldSG with instance mask · 3DSSG | Relationship recall: 40.3 -> 47.8 | Improved |
| add dual-domain depth refinement (DR) (ablation) | DeWorldSG with instance mask · 3DSSG | Predicate recall: 46.7 -> 52.9 | Improved |
| add V-JEPA 2 world-model prior (ablation) | DeWorldSG with mask and DR · 3DSSG | Relationship recall: 47.8 -> 50.2 | Improved |
| add V-JEPA 2 world-model prior (ablation) | DeWorldSG with mask and DR · 3DSSG | Predicate mRecall: 30.1 -> 36.0 | Improved |
| add V-JEPA 2 world-model prior (ablation) | DeWorldSG with mask and DR · 3DSSG | Object mRecall: 70.8 -> 70.6 | No clear effect |
| use V-JEPA 2 temporal prior instead of DINOv2 static prior (ablation) | DeWorldSG · 3DSSG | Relationship recall: 49.5 -> 50.2 | Improved |
| use V-JEPA 2 temporal prior instead of DINOv2 static prior (ablation) | DeWorldSG · 3DSSG | Predicate mRecall: 33.7 -> 36.0 | Improved |
| add DINOv2 prior instead of no prior (ablation) | DeWorldSG with mask and DR · 3DSSG | Predicate recall: 52.9 -> 56.6 | Improved |
| replace GT camera poses with ORB-SLAM3 estimated poses | DeWorldSG · ReplicaSSG | Relationship recall: 38.2 -> 37.4 | Hurt |
| replace GT camera poses with ORB-SLAM3 estimated poses | DeWorldSG · ReplicaSSG | Object recall: 35.4 -> 33.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FROSS — FROSS: Faster-Than-Real-Time Online 3D Semantic Scene Graph Generation from RGB-D Images (ICCV 2025) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GraphWrit3R: End-to-End Writing of Scene Graphs for Multi-Modal 3D Scenes (NeurIPS 2026 · similarity 0.92)
- OP3DSG: Open-vocabulary Part-aware 3D Scene Graph Generation for Real-world Environments (ECCV 2026 · similarity 0.91)
- FROSS: Faster-Than-Real-Time Online 3D Semantic Scene Graph Generation from RGB-D Images (ICCV 2025 · similarity 0.91)
- Open3DSG: Open-Vocabulary 3D Scene Graphs from Point Clouds with Queryable Objects and Open-Set Relationships (CVPR 2024 · similarity 0.90)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.90)
- GraphDreamer: Compositional 3D Scene Synthesis from Scene Graphs (CVPR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
