Research map / 3D vision / 3D generation
Learning 3D Scene Analogies with Neural Contextual Scene Maps
Junho Kim, Gwangtak Bae, Eun Sun Lee, Young Min Kim
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 local displacement estimation after affine map (ablation) | neural contextual scene maps · 3D-FRONT (avg of manual and procedural pairs) | Bijectivity PCP@0.25: 0.83 -> 0.90 | Improved |
| add local displacement estimation after affine map (ablation) | neural contextual scene maps · 3D-FRONT (avg of manual and procedural pairs) | Chamfer Acc.@0.15: 0.77 -> 0.94 | Improved |
| add semantic embedding to descriptor field (ablation) | neural contextual scene maps · 3D-FRONT manual pairs (embedding ablation) | Bijectivity PCP@0.25: 0.50 -> 0.70 | Improved |
| add distance embedding to descriptor field (ablation) | neural contextual scene maps · 3D-FRONT manual pairs (embedding ablation) | Bijectivity PCP@0.25: 0.67 -> 0.70 | Improved |
| use CLIP embedding instead of semantic label embedding (ablation) | neural contextual scene maps · 3D-FRONT (avg of manual and procedural pairs) | Bijectivity PCP@0.25: 0.90 -> 0.77 | Mixed |
| use sentence embedding instead of semantic label embedding (ablation) | neural contextual scene maps · 3D-FRONT (avg of manual and procedural pairs) | Chamfer Acc.@0.20: 0.96 -> 0.97 | No clear effect |
| use learned descriptor fields instead of Vector Neuron 3D point features | coarse-to-fine map estimation · 3D-FRONT procedural pairs | PCP@0.25: 0.56 -> 0.76 | Improved |
| use learned descriptor fields instead of DINOv2 visual features | coarse-to-fine map estimation · 3D-FRONT manual pairs | Bijectivity PCP@0.25: 0.56 -> 0.70 | Improved |
| use MAE instead of DeiT III features in visual feature field baseline | Visual Feature Field · 3D-FRONT manual pairs | Bijectivity PCP@0.25: 0.31 -> 0.50 | Improved |
| increase RoI points sampled per object from 50 to 400 | neural contextual scene maps · 3D-FRONT procedural pairs | PCP@0.25: 0.68 -> 0.76 | Improved |
| different instead of identical object group cardinality | neural contextual scene maps · 3D-FRONT manual pairs | Chamfer Acc.@0.15: 0.76 -> 0.66 | Hurt |
| our method versus 3D point feature field in sim-to-real pairs | neural contextual scene maps · Sim2Real manual pairs (102 pairs) | Chamfer Acc.@0.15: 0.55 -> 0.66 | Improved |
Closest papers by meaning
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.87)
- Uncovering and Shaping the Latent Representation of 3D Scene Topology in Vision-Language Models (NeurIPS 2026 · similarity 0.87)
- DORSal: Diffusion for Object-centric Representations of Scenes $\textit{et al.}$ (ICLR 2024 · similarity 0.87)
- LSceneLLM: Enhancing Large 3D Scene Understanding Using Adaptive Visual Preferences (CVPR 2025 · similarity 0.87)
- 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning (CVPR 2025 · similarity 0.86)
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · 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.
