Research map / 3D vision / 3D generation
RAG-3DSG: Enhancing 3D Scene Graphs with Re-Shot Guided Retrieval-Augmented Generation
Yue Chang, Rufeng Chen, Zhaofan ZHANG, Yi Chen, Yifan Tian, Sihong Xie
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 re-shot guided uncertainty estimation (ablation) | RAG-3DSG · Replica ablation | mF1: 14.66 vs. 30.78 | Improved |
| add re-shot guided uncertainty estimation (ablation) | RAG-3DSG · Replica ablation | f-mIoU: 35.56 vs. 54.26 | Improved |
| add uncertainty-guided filtering before RAG (ablation) | RAG-3DSG · Replica ablation | mF1: 26.52 -> 30.78 | Improved |
| retrieve top-1 anchor instead of top-3 (ablation) | RAG-3DSG · Replica ablation | mF1: 28.26 -> 30.78 | Improved |
| retrieve top-1 anchor instead of top-5 (ablation) | RAG-3DSG · Replica ablation | mF1: 27.66 -> 30.78 | Improved |
| add object-level RAG and image concatenation to re-shot caption (ablation) | RAG-3DSG (GPT-4o) · SceneFun3D | Object R@3: 59.4 -> 83.0 | Improved |
| add active re-shot verification over passive crops (ablation) | RAG-3DSG (GPT-4o) · SceneFun3D | Object R@3: 67.7 -> 83.0 | Improved |
| add Gaussian noise to camera poses | Ours-GPT · SceneFun3D | Object R@3: 83.0 -> 82.1 | No clear effect |
| replace ConceptGraphs edge generator with ours (ablation) | Ours nodes · SceneFun3D | Edge R@10: 95.6 -> 97.9 | Improved |
| reduce point density by 50% (ablation) | RAG-3DSG · Replica | mF1: -3.48% | Hurt |
| increase point density by 50% (ablation) | RAG-3DSG · Replica | mF1: +0.07% | No clear effect |
| use dynamic downsampling instead of ConceptGraphs fusion | RAG-3DSG · Replica | per-iteration time: 6.65s to 2.49s | Improved |
Closest papers by meaning
- DepWorldSG: Depth-Aware 3D Semantic Scene Graph Generation via World-Model Priors (ECCV 2026 · similarity 0.90)
- Open-World 3D Scene Graph Generation for Retrieval-Augmented Reasoning (AAAI 2026 · similarity 0.89)
- OP3DSG: Open-vocabulary Part-aware 3D Scene Graph Generation for Real-world Environments (ECCV 2026 · similarity 0.89)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.88)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.87)
- Tackling View-Dependent Semantics in 3D Language Gaussian Splatting (ICML 2025 · similarity 0.87)
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.
