Research map / 3D vision / 3D generation
Understanding Real World Indoor Scenes With Synthetic Data
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 |
|---|---|---|---|
| Add synthetic SceneNet pretraining before fine-tuning on NYUv2 | DHA depth segmentation model · NYUv2 test, 11 classes | class accuracy: 53.7 -> 59.1 | Helped |
| Add synthetic SceneNet pretraining before fine-tuning on NYUv2 | DHA depth segmentation model · NYUv2 test, 11 classes | global accuracy: 63.8 -> 67.4 | Helped |
| Use synthetic SceneNet training only, without real-data fine-tuning (ablation) | DHA depth segmentation model · NYUv2 test, 11 classes | class accuracy: 53.7 -> 42.6 | Hurt |
| Add synthetic SceneNet pretraining before fine-tuning on NYUv2 | DHA depth segmentation model · NYUv2 test, 13 classes | class accuracy: 47.2 -> 51.7 | Helped |
| Add synthetic SceneNet pretraining before fine-tuning on SUN RGB-D | DHA depth segmentation model · SUN RGB-D test, 13 classes | class accuracy: 49.2 -> 52.7 | Helped |
| Add dropout to NYU-only training (ablation) | DHA depth segmentation model · NYUv2 test, 11 classes | class accuracy: 53.7 -> 56.9 | Helped |
| Add dropout to SceneNet pretraining followed by NYUv2 fine-tuning (ablation) | DHA depth segmentation model · NYUv2 test, 11 classes | class accuracy: 59.1 -> 59.9 | Helped |
| Add dropout to SceneNet pretraining followed by NYUv2 fine-tuning (ablation) | DHA depth segmentation model · NYUv2 test, 13 classes | class accuracy: 51.7 -> 52.5 | Helped |
| Add dropout to SceneNet pretraining followed by SUN RGB-D fine-tuning (ablation) | DHA depth segmentation model · SUN RGB-D test, 13 classes | class accuracy: 52.7 -> 53.1 | No clear effect |
| Add synthetic pretraining to SUNRGBD-DO-DHA | DHA depth segmentation model with dropout · SUN RGB-D test, 13 classes | global accuracy: 74.2 -> 75.0 | Helped |
| Use synthetic SceneNet training only, without real-data fine-tuning (ablation) | DHA depth segmentation model · SUN RGB-D test, 13 classes | class accuracy: 49.2 -> 30.2 | Hurt |
| Add synthetic pretraining and real-data fine-tuning, using depth only instead of RGBD and normals | SceneNet-FT-NYU-DHA versus Eigen et al. · NYUv2 test, 11 classes | global accuracy: 69.5 vs 67.4 | Mixed |
Closest papers by meaning
- Physically-Based Rendering for Indoor Scene Understanding Using Convolutional Neural Networks (CVPR 2017 · similarity 0.91)
- ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes (CVPR 2017 · similarity 0.88)
- ARKit LabelMaker: A New Scale for Indoor 3D Scene Understanding (CVPR 2025 · similarity 0.86)
- SceneCraft: Layout-Guided 3D Scene Generation (NeurIPS 2024 · similarity 0.86)
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 2024 · similarity 0.85)
- SceneFun3D: Fine-Grained Functionality and Affordance Understanding in 3D Scenes (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.