Research map / 3D vision / 3D generation
SceneGraphNet: Neural Message Passing for 3D Indoor Scene Augmentation
Yang Zhou, Zachary While, Evangelos Kalogerakis
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 dense co-occurrence edges to the scene graph (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 60.0 -> 67.3 | Improved |
| use full graph with all relationships instead of tree structure (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 61.1 -> 67.3 | Improved |
| add supporting/surrounding relationships to co-occurrence-only graph (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 56.5 -> 67.3 | Improved |
| use GRU aggregation instead of summation (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 57.7 -> 67.3 | Improved |
| use GRU aggregation instead of max-pooling (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 63.1 -> 67.3 | Improved |
| use GRU aggregation instead of vanilla RNN (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 64.8 -> 67.3 | Improved |
| add learned attention on messages (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 60.3 -> 67.3 | Improved |
| use learned attention instead of distance-based weights (ablation) | SceneGraphNet · SUNCG scene augmentation | top-1 accuracy: 63.8 -> 67.3 | Improved |
| run second message passing iteration instead of one (ablation) | SceneGraphNet · SUNCG scene augmentation | top-5 accuracy: 72.7 -> 88.2 | Improved |
| run third message passing iteration instead of two (ablation) | SceneGraphNet · SUNCG scene augmentation | top-5 accuracy: 88.2 -> 89.5 | Improved |
| combine SceneGraphNet posterior with MVCNN (ablation) | MVCNN · SUNCG object recognition | accuracy: 59.2 -> 72.2 | Improved |
| use SceneGraphNet instead of GRAINS for size prediction (ablation) | scene augmentation · SUNCG size prediction | size error (cm): 38 -> 26 | Improved |
Closest papers by meaning
- Learning 3D Semantic Scene Graphs From 3D Indoor Reconstructions (CVPR 2020 · similarity 0.90)
- Learning 3D Semantic Scene Graphs with Instance Embeddings (IJCV 2022 · similarity 0.88)
- Open-Vocabulary Functional 3D Scene Graphs for Real-World Indoor Spaces (CVPR 2025 · similarity 0.87)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.86)
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 2024 · similarity 0.86)
- GraphDreamer: Compositional 3D Scene Synthesis from Scene Graphs (CVPR 2024 · similarity 0.85)
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.
