Research map / Probabilistic and scientific modeling / Variational and generative sampling
SG-VAE: Scene Grammar Variational Autoencoder to generate new indoor scenes
Pulak Purkait, Christopher Zach, Ian Reid
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 |
|---|---|---|---|
| use rule-concatenated attributes (SG-VAE) instead of per-rule zero-filled attributes (BL1) (ablation) | SG-VAE · SUN RGB-D room layout | IoU: 0.5673 -> 0.6240 | Improved |
| add grammar and relative pose (SG-VAE vs No Grammar VAE BL2) (ablation) | SG-VAE · SUN RGB-D room layout | IoU: 0.2964 -> 0.6240 | Improved |
| learn pose and shape jointly in the autoencoder (SG-VAE vs Grammar VAE + Make home BL3) (ablation) | SG-VAE · SUN RGB-D room layout | IoU: 0.5119 -> 0.6240 | Improved |
| add grammar (SG-VAE vs No Grammar VAE) on chair reconstruction (ablation) | SG-VAE · SUN RGB-D chair boxes | detection rate: 33.7 -> 92.3 | Improved |
| use SG-VAE attribute design instead of BL1 on piano (ablation) | SG-VAE · SUN RGB-D piano boxes | detection rate: 48.7 -> 67.1 | Improved |
| use SG-VAE instead of BL1 on kitchen counter (ablation) | SG-VAE · SUN RGB-D kitchen counter boxes | detection rate: 100.0 -> 100.0 | No clear effect |
| use SG-VAE latent to rescore DSS detections vs DSS alone | SG-VAE + DSS · SUN RGB-D RGB-D to room layout | IoU: 0.4070 -> 0.4387 | Improved |
| use SG-VAE vs BL1 for RGB-D layout (ablation) | SG-VAE · SUN RGB-D RGB-D to room layout | IoU: 0.4315 -> 0.4387 | Improved |
| use SG-VAE instead of Grains on SUNCG scene realism | SG-VAE · SUNCG classifier | accuracy (lower better): 96.4% -> 83.7% | Improved |
Closest papers by meaning
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 2024 · similarity 0.86)
- Hierarchically-Structured Open-Vocabulary Indoor Scene Synthesis with Pre-trained Large Language Model (AAAI 2025 · similarity 0.85)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · similarity 0.85)
- Scene Graph Disentanglement and Composition for Generalizable Complex Image Generation (NeurIPS 2024 · similarity 0.84)
- S-INF: Towards Realistic Indoor Scene Synthesis via Scene Implicit Neural Field (AAAI 2025 · similarity 0.84)
- SparseGNV: Generating Novel Views of Indoor Scenes with Sparse RGB-D Images (AAAI 2024 · similarity 0.84)
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.
