Research map / 3D vision / 3D generation
Configurable 3D Scene Synthesis and 2D Image Rendering with Per-pixel Ground Truth Using Stochastic Grammars
Chenfanfu Jiang, Siyuan Qi, Yixin Zhu, Siyuan Huang, Jenny Lin, Lap-Fai Yu, Demetri Terzopoulos, Song Chun Zhu
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 |
|---|---|---|---|
| pretrain on synthetic data before fine-tuning on NYUv2 | FCN of [146] · NYU-Depth V2 normal estimation | mean angular error: 21.74 -> 21.47 | Improved |
| pretrain on synthetic data before fine-tuning on NYUv2 | FCN of [146] · NYU-Depth V2 normal estimation | accuracy within 11.25 degrees: 39.37 -> 39.84 | Improved |
| pretrain on synthetic data then fine-tune on NYUv2 instead of training only on NYUv2 | network of [71] · NYU-Depth V2 depth estimation | Abs Rel: 0.233 -> 0.226 | Improved |
| train on synthetic data only instead of NYUv2 only | network of [71] · NYU-Depth V2 depth estimation | Abs Rel: 0.233 -> 0.241 | Hurt |
| train on synthetic data only instead of NYUv2 only | network of [71] · NYU-Depth V2 depth estimation | accuracy delta<1.25: 0.605 -> 0.612 | Improved |
| lower noise-level threshold to higher (0.1 instead of 0.001) (ablation) | Mantra renderer · Rendering time vs quality | render time (seconds): 203 -> 30 | Improved |
| reduce baseline pixel samples from 3x3 to 1x1 (ablation) | Mantra renderer · Rendering time vs quality | render time (seconds): 203 -> 45 | Improved |
| reduce max additional rays from 22 to 3 (ablation) | Mantra renderer · Rendering time vs quality | render time (seconds): 203 -> 36 | Improved |
| change illumination intensity of test scenes | depth method [29] · Synthetic benchmark depth estimation | Abs Rel: 0.373 -> 0.483 | Hurt |
| change illumination color of test scenes | depth method [71] · Synthetic benchmark depth estimation | Abs Rel: 0.225 -> 0.332 | Hurt |
| change object material properties of test scenes | depth method [71] · Synthetic benchmark depth estimation | Abs Rel: 0.225 -> 0.192 | No clear effect |
| change illumination color of test scenes | normal method [28] · Synthetic benchmark normal estimation | mean angular error: 22.74 -> 26.53 | Mixed |
Closest papers by meaning
- Human-Centric Indoor Scene Synthesis Using Stochastic Grammar (CVPR 2018 · similarity 0.89)
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 2024 · similarity 0.89)
- SceneWeaver: All-in-One 3D Scene Synthesis with an Extensible and Self-Reflective Agent (NeurIPS 2025 · similarity 0.88)
- Video Perception Models for 3D Scene Synthesis (NeurIPS 2025 · similarity 0.88)
- Architect: Generating Vivid and Interactive 3D Scenes with Hierarchical 2D Inpainting (NeurIPS 2024 · similarity 0.87)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.87)
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.
