Research map / 3D vision / Depth and stereo
Self-Training Room Layout via Geometry-aware Ray-casting
Bolivar Solarte, Chin-Hsuan Wu, Jin-Cheng Jhang, Jonathan Lee, Yi-Hsuan Tsai, Min Sun
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 |
|---|---|---|---|
| replace 360-MLC pseudo-labels with ray-casting pseudo-labels | HorizonNet self-training · HM3D-MVL occlusion subset (100% data) | 2D IoU (%): 79.19 -> 83.01 | Improved |
| increase self-training data from 10% to 100% with 360-MLC pseudo-labels | 360-MLC · HM3D-MVL occlusion subset | 2D IoU (%): 81.66 -> 79.19 | Hurt |
| increase self-training data from 10% to 100% with ray-casting pseudo-labels | Ours (HorizonNet) · HM3D-MVL occlusion subset | 2D IoU (%): 82.05 -> 83.01 | Improved |
| swap 360-MLC for ray-casting pseudo-labels with LGTNet backbone | LGTNet self-training · HM3D-MVL occlusion subset | 2D IoU (%): 71.29 -> 83.75 | Improved |
| self-train with ray-casting pseudo-labels from pre-trained model | LGTNet · ZInD test | 2D IoU (%): 72.59 -> 76.77 | Improved |
| use ray-casting pseudo-labels without any weighting (ablation) | HorizonNet self-training · HM3D-MVL test (10% data) | 2D IoU (%): 76.71 -> 81.65 | Improved |
| add uncertainty-only weighting sigma^-2 (ablation) | ray-casting pseudo-labels · HM3D-MVL (10% data) | 2D IoU (%): 81.65 -> 81.02 (full test); occlusion 80.85 -> 81.28 | Mixed |
| add weighted-distance loss prioritizing farthest geometry (ablation) | ray-casting pseudo-labels · HM3D-MVL (10% data) | 3D IoU (%): 76.99 -> 77.99 | Improved |
| add weighted-distance loss versus unweighted pseudo-labels on occlusion subset (ablation) | ray-casting pseudo-labels · HM3D-MVL occlusion subset (10% data) | 3D IoU (%): 78.98 -> 80.45 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 360-MLC — 360-MLC: Multi-view Layout Consistency for Self-training and Hyper-parameter Tuning (NeurIPS 2022) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 360-MLC: Multi-view Layout Consistency for Self-training and Hyper-parameter Tuning (NeurIPS 2022 · similarity 0.91)
- PolyLayout: Multi-room Manhattan Layout Estimation (ECCV 2026 · similarity 0.90)
- HouseLayout3D: A Benchmark and Training-free Baseline for 3D Layout Estimation in the Wild (NeurIPS 2025 · similarity 0.87)
- OptiScene: LLM-driven Indoor Scene Layout Generation via Scaled Human-aligned Data Synthesis and Multi-Stage Preference Optimization (NeurIPS 2025 · similarity 0.87)
- No More Ambiguity in 360deg Room Layout via Bi-Layout Estimation (CVPR 2024 · similarity 0.87)
- Direct Numerical Layout Generation for 3D Indoor Scene Synthesis via Spatial Reasoning (NeurIPS 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.
