Research map / 3D vision / 3D generation
SceneGraphFusion: Incremental 3D Scene Graph Prediction From RGB-D Sequences
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use SceneGraphFusion rather than 3DSSG | 3D scene graph prediction with ground-truth instances · 3RScan/3DSSG | Relationship R@50: 0.40 -> 0.85 | Helped |
| Use feature-wise attention (FAT) rather than no attention (ablation) | SceneGraphFusion · 3RScan/3DSSG geometric segments, full scene | Relationship R@1: 0.41 -> 0.55 | Helped |
| Use feature-wise attention (FAT) rather than GAT (ablation) | SceneGraphFusion · 3RScan/3DSSG geometric segments, full scene | Relationship R@1: 0.12 -> 0.55 | Helped |
| Use feature-wise attention (FAT) rather than SDPA (ablation) | SceneGraphFusion · 3RScan/3DSSG geometric segments, full scene | Relationship R@1: 0.39 -> 0.55 | Helped |
| Use feature-wise attention rather than no attention (ablation) | SceneGraphFusion · 3RScan segment classification with all edges | avg. IoU: 33.5 -> 49.3 | Helped |
| Use feature-wise attention rather than no attention with 50% edge drop (ablation) | SceneGraphFusion · 3RScan segment classification with 50% edge drop | avg. IoU: 29.5 -> 41.9 | Helped |
| Train with predicate loss jointly with object loss (ablation) | SceneGraphFusion · 3RScan/3DSSG geometric segments, full scene | Object R@1: 0.62 -> 0.75 | Helped |
| Add temporal prediction fusion (ablation) | SceneGraphFusion panoptic segmentation · ScanNet v2 open test set | PQ: 35.4 -> 36.3 | Helped |
| Add the same part relationship (ablation) | SceneGraphFusion panoptic segmentation · ScanNet v2 open test set | PQ: 10.9 -> 36.3 | Helped |
| Use incremental prediction fusion instead of incremental prediction without fusion | SceneGraphFusion · 3RScan/3DSSG geometric segments, incremental | Relationship R@1: 0.51 -> 0.52 | Helped |
| Ignore missing reconstructed regions during evaluation rather than map them to nearest neighbors | SceneGraphFusion panoptic segmentation · ScanNet v2 open test set | PQ, all: 31.5 -> 36.3 | Helped |
| Use SceneGraphFusion rather than PanopticFusion | Panoptic segmentation with nearest-neighbor mapping · ScanNet v2 open test set | PQ, all: 33.5 -> 31.5 | Hurt |
Closest papers by meaning
- Open3DSG: Open-Vocabulary 3D Scene Graphs from Point Clouds with Queryable Objects and Open-Set Relationships (CVPR 2024 · similarity 0.88)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.88)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.87)
- Adaptive Visual Scene Understanding: Incremental Scene Graph Generation (NeurIPS 2024 · similarity 0.87)
- Open-Vocabulary Functional 3D Scene Graphs for Real-World Indoor Spaces (CVPR 2025 · similarity 0.87)
- SGFormer: Semantic Graph Transformer for Point Cloud-Based 3D Scene Graph Generation (AAAI 2024 · 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.