Research map / 3D vision / 3D generation
Generating Holistic 3D Scene Abstractions for Text-Based Image Retrieval
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 |
|---|---|---|---|
| Generate holistic 3D layouts from text, project them to 2D, and match detected boxes | Text-based image retrieval · SUN RGB-D | R@10: 15.3 -> 18.7 | Helped |
| Combine normalized retrieval scores with the learned 2D baseline | Ours soft[5,5] · SUN RGB-D | R@1: 4.9 -> 8.7 | Helped |
| Use confidence-weighted detections rather than thresholded boxes with uniform weights (ablation) | Ours hard[5,1] · SUN RGB-D | R@1: 3.9 -> 4.5 | Helped |
| Use confidence-weighted detections rather than thresholded boxes with uniform weights (ablation) | Ours hard[5,1] · SUN RGB-D | R@10: 16.4 -> 16.7 | Helped |
| Increase camera views per layout from 1 to 5 (ablation) | Ours soft · SUN RGB-D | R@1: 4.5 -> 4.9 | Helped |
| Increase camera views per layout from 1 to 5 (ablation) | Ours soft · SUN RGB-D | R@10: 16.7 -> 18.7 | Helped |
| Increase the number of sampled 3D layouts (ablation) | Ours · SUN RGB-D | median rank | Helped |
| Increase the number of sampled camera views (ablation) | Ours · SUN RGB-D | median rank | Helped |
| Use 3D layout matching with DPM detections | Text-based image retrieval · 3DGP w/ DPM bbox | R@100: 46.0 -> 52.0 | Helped |
| Use 3D layout matching with ground-truth boxes | Text-based image retrieval · 3DGP w/ GT bbox | R@50: 38.0 -> 41.4 | Helped |
| Use 3D layout matching with ground-truth boxes | Text-based image retrieval · 3DGP w/ GT bbox | R@1: 4.0 -> 3.0 | Hurt |
| Apply interval shrinkage during feasible-layout search (ablation) | 3D scene solver · Example query from Table 1 | solver time: 10 minutes -> 6 seconds | Helped |
Closest papers by meaning
- Disentangled 3D Scene Generation with Layout Learning (ICML 2024 · similarity 0.87)
- GraphDreamer: Compositional 3D Scene Synthesis from Scene Graphs (CVPR 2024 · similarity 0.87)
- SceneForge: Enhancing 3D-text alignment with Structured Scene Compositions (NeurIPS 2025 · similarity 0.86)
- Open-Vocabulary Functional 3D Scene Graphs for Real-World Indoor Spaces (CVPR 2025 · similarity 0.86)
- RelationVGGT : Visual Geometry Transformers for 3D Spatial Relation Segmentation (NeurIPS 2026 · similarity 0.86)
- SceneScaffold: Active Scene-State Construction for Unified 3D Scene Understanding (NeurIPS 2026 · similarity 0.86)
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.