Research map / Image generation / Text-to-image diffusion
SceneComposer: Any-Level Semantic Image Synthesis
Yu Zeng, Zhe Lin, Jianming Zhang, Qing Liu, John Collomosse, Jason Kuen, Vishal M. Patel
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 the unified any-level model instead of the text-to-image-only baseline (ablation) | SceneComposer · OpenLayout-COCO, precision 0 | FID: 32.83 -> 32.09 | Improved |
| Use the unified any-level model instead of the fixed-level layout baseline (ablation) | SceneComposer · OpenLayout-COCO, precision 4 | SS Score: .5729 -> .6978 | Improved |
| Increase layout precision from 0 to 6 | SceneComposer · OpenLayout | SS Score: .436 -> .558 | Improved |
| Increase layout precision from 0 to 6 | SceneComposer · OpenLayout-COCO | SS Score: .572 -> .736 | Improved |
| Increase layout precision from 0 to 6 | SceneComposer · OpenLayout-COCO | FID: 25.3 -> 30.6 | Hurt |
| Increase layout precision from 0 to 6 | SceneComposer · OpenLayout | SS Score: .436 -> .558 | Mixed |
| Use the any-level model instead of the text-to-image-only baseline (ablation) | SceneComposer · OpenLayout-COCO, precision 0 | CLIP Score: .2473 -> .2496 | Improved |
| Use the any-level model instead of the fixed-level layout baseline (ablation) | SceneComposer · OpenLayout-COCO, precision 4 | FID: 36.03 -> 33.70 | Improved |
| Use classifier-free guidance with all text feature pyramid levels (ablation) | SceneComposer · COCO-stuff val | SS Score: .733 -> .736 | Improved |
| Randomly permute the order of regional text labels (ablation) | SceneComposer · COCO-stuff val | FID: 17.20 -> 17.21 | No clear effect |
| Randomly permute the order of regional text labels (ablation) | SceneComposer · COCO-stuff val | mIOU: 23.01 -> 22.76 | No clear effect |
| Use SceneComposer rather than VQ-Diffusion | COCO text-to-image generation · COCO val | FID: 13.86 -> 8.55 | Improved |
Closest papers by meaning
- Inferring Semantic Layout for Hierarchical Text-to-Image Synthesis (CVPR 2018 · similarity 0.90)
- PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis (CVPR 2024 · similarity 0.89)
- Training-free Dense-Aligned Diffusion Guidance for Modular Conditional Image Synthesis (CVPR 2025 · similarity 0.87)
- SpotActor: Training-Free Layout-Controlled Consistent Image Generation (AAAI 2025 · similarity 0.87)
- Open Vocabulary Semantic Scene Sketch Understanding (CVPR 2024 · similarity 0.86)
- LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration (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.
