Research map / Image generation / Text-to-image diffusion
Laytrol: Preserving Pretrained Knowledge in Layout Control for Multimodal Diffusion Transformers
Sida Huang, Siqi Huang, Ping Luo, Hongyuan Zhang
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 |
|---|---|---|---|
| add parameter copy initialization (ablation) | Laytrol · COCO 2017 (200k iterations) | mIoU: 64.92 -> 75.68 | Improved |
| add parameter copy initialization (ablation) | Laytrol · COCO 2017 (200k iterations) | FID: 38.22 -> 35.65 | Improved |
| add layout-level RoPE (ablation) | Laytrol · COCO 2017 (200k iterations) | mIoU: 64.92 -> 67.02 | Improved |
| add random prompt dropping (ablation) | Laytrol · COCO 2017 (200k iterations) | mIoU: 64.92 -> 71.28 | Improved |
| add random prompt dropping (ablation) | Laytrol · COCO 2017 (200k iterations) | FID: 38.22 -> 36.70 | Improved |
| combine all three components (full model) (ablation) | Laytrol · COCO 2017 (200k iterations) | AP: 51.78 -> 64.11 | Improved |
| control every MM-DiT block instead of every sixth (interval 6 to 1) (ablation) | Laytrol · COCO 2017 | mIoU: 72.16 -> 76.75 | Improved |
| control every MM-DiT block instead of every sixth (interval 6 to 1) (ablation) | Laytrol · COCO 2017 | FID: 42.48 -> 35.38 | Improved |
| control every MM-DiT block (Laytrol-1) vs FLUX (ablation) | Laytrol · Inference cost | latency: 2.9s -> 6.4s | Hurt |
| increase interval from 1 to 6 (Laytrol-6) (ablation) | Laytrol · Inference cost | latency: 6.4s -> 4.3s | Improved |
| Laytrol vs SiamLayout-FLUX on spatial relation | Laytrol · T2I-CompBench | Spatial: 35.84 -> 47.40 | Improved |
| Laytrol vs FLUX.1 on non-spatial | Laytrol · T2I-CompBench | Non-Spatial: 31.27 -> 30.40 | No clear effect |
Closest papers by meaning
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.88)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.88)
- Build-A-Scene: Interactive 3D Layout Control for Diffusion-Based Image Generation (ICLR 2025 · similarity 0.87)
- Control and Realism: Best of Both Worlds in Layout-to-Image without Training (ICML 2025 · similarity 0.87)
- GrounDiT: Grounding Diffusion Transformers via Noisy Patch Transplantation (NeurIPS 2024 · similarity 0.87)
- Adversarial Supervision Makes Layout-to-Image Diffusion Models Thrive (ICLR 2024 · 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.
