Research map / 3D vision / 3D generation
3D Scene Graph: A Structure for Unified Semantics, 3D Space, and Camera
Iro Armeni, Zhi-Yang He, JunYoung Gwak, Amir R. Zamir, Martin Fischer, Jitendra Malik, Silvio Savarese
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 framing of query images sampled on panoramas (ablation) | Mask R-CNN · Gibson 2D panoramas | AP: 0.079 -> 0.160 | Improved |
| add multi-view consistency on top of framing (ablation) | Mask R-CNN + Framing · Gibson 2D panoramas | AP: 0.160 -> 0.485 | Improved |
| add framing before panorama projection (ablation) | Mask R-CNN + Pano Projection · Gibson 3D mesh | AP: 0.222 -> 0.306 | Improved |
| add multi-view consistency on top of framing (ablation) | Mask R-CNN + Framing + Pano Projection · Gibson 3D mesh | AP: 0.306 -> 0.409 | Improved |
| add framing to a weaker detector (ablation) | BlitzNet · Gibson 2D panoramas | AP: 0.095 -> 0.198 | Improved |
| add multi-view consistency to a weaker detector (ablation) | BlitzNet + Framing · Gibson 2D panoramas | AP: 0.198 -> 0.284 | Improved |
| add framing and MVC to a weaker detector (ablation) | BlitzNet · Gibson 3D mesh | AP: 0.076 -> 0.245 | Improved |
| add human manual verification to the automatic pipeline | Ours (fully automatic) · Gibson tiny split | AP: 0.389 -> 0.97 | Improved |
| use automatic pipeline plus verification instead of fully manual 2D annotation | Ours w/ human (MV) · Gibson annotation time | hours: 12:44:10 -> 03:18:02 | Improved |
| train on 3D-scene-graph-derived masks for spatial order | ResNet34 · Spatial order classification | mean AP: 0.255 -> 0.712 | Improved |
| train on 3D-scene-graph-derived masks for relative volume | ResNet34 · Relative volume classification | mean AP: 0.555 -> 0.820 | Improved |
| learn amodal mask prediction with U-Net instead of class-specific average mask | U-Net · Amodal mask segmentation | avg f1-score: 0.545 -> 0.672 | Improved |
Closest papers by meaning
- Learning 3D Semantic Scene Graphs From 3D Indoor Reconstructions (CVPR 2020 · similarity 0.92)
- Beyond Isolated Objects: Relationship-aware Open Vocabulary Scene Understanding via 3D Scene Graph Analysis (ECCV 2026 · similarity 0.91)
- The Scene Language: Representing Scenes with Programs, Words, and Embeddings (CVPR 2025 · similarity 0.90)
- Open-Vocabulary Functional 3D Scene Graphs for Real-World Indoor Spaces (CVPR 2025 · similarity 0.89)
- SceneFun3D: Fine-Grained Functionality and Affordance Understanding in 3D Scenes (CVPR 2024 · similarity 0.89)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · 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.
