Research map / Image generation / Text-to-image diffusion
Lay-Your-Scene: Natural Scene Layout Generation with Diffusion Transformers
Divyansh Srivastava, Xiang Zhang, He Wen, Chenru Wen, Zhuowen Tu
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 LayouSyn diffusion layout generator instead of LayoutGPT (GPT-4o-mini) | LayoutGPT · COCO-GR validation | L-FID: 6.72 -> 3.07 | Improved |
| add GRIT pretraining (ablation) | LayouSyn · COCO-GR validation | L-FID: 3.07 -> 3.31 | Hurt |
| add GRIT pretraining (ablation) | LayouSyn · NSR-1K spatial reasoning | accuracy: 87.49 -> 92.58 | Improved |
| add GRIT pretraining (ablation) | LayouSyn · NSR-1K spatial reasoning | GLIP accuracy: 54.91 -> 58.94 | Improved |
| use LayouSyn instead of LayoutGPT with fixed GPT-4o-mini description set | LayoutGPT · COCO-GR L-FID | L-FID: 6.72 -> 3.32 | Improved |
| generate description set with Llama-3.1-8B instead of GPT-3.5 (ablation) | LayouSyn · COCO-GR L-FID | L-FID: 3.49 -> 2.74 | Improved |
| increase DDIM sampling steps from 10 to 100 (ablation) | LayouSyn · COCO-GR L-FID | L-FID: 3.47 -> 2.74 | Improved |
| add cross-attention between prompt and description tokens (ablation) | LDiT block · COCO-GR L-FID | L-FID: 2.82 -> 2.81 | No clear effect |
| add modulation of description tokens on top of cross-attention (ablation) | LDiT block · COCO-GR L-FID | L-FID: 2.81 -> 2.74 | Improved |
| increase noise schedule scale s to 2.0 with CFG 2.0 (ablation) | LayouSyn · COCO-GR L-FID | L-FID | Improved |
| initialize with DDIM inversion of LLM layout plus description set (vs description set only) | LayouSyn · NSR-1K spatial reasoning (Llama-3.1-8B) | accuracy: 89.75 -> 90.46 | Improved |
| use LayouSyn description set + inversion on GPT-4o-mini layouts | LayoutGPT (GPT-4o-mini) · NSR-1K spatial reasoning | accuracy: 92.01 -> 92.08 | No clear effect |
Closest papers by meaning
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.89)
- Build-A-Scene: Interactive 3D Layout Control for Diffusion-Based Image Generation (ICLR 2025 · similarity 0.88)
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.88)
- SceneFactor: Factored Latent 3D Diffusion for Controllable 3D Scene Generation (CVPR 2025 · similarity 0.88)
- CoT-lized Diffusion: Let's Reinforce T2I Generation Step-by-step (NeurIPS 2025 · similarity 0.88)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.88)
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.
