Research map / Image generation / Text-to-image diffusion
Layout-Conditioned Autoregressive Text-to-Image Generation via Structured Masking
Zirui Zheng, Takashi Isobe, Tong Shen, XU JIA, Xiaomin Li, Jianbin Zhao, Mengmeng Ge, Baolu Li, Qinghe Wang, Haiwen Diao, Dong Li, Yunzhi Zhuge and 3 more
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 GRPO post-training with HPS plus layout reward (ablation) | SMARLI-SFT · LayoutSAM-Eval | ImageReward / mIoU: IR 60.88 -> 74.81; mIoU 80.08 -> 81.31 | Improved |
| use HPS reward alone in GRPO (ablation) | SMARLI-SFT · LayoutSAM-Eval | Color / mIoU: Color 86.17 -> 83.96; mIoU 80.08 -> 77.79 | Hurt |
| use layout reward alone in GRPO (ablation) | SMARLI-SFT · LayoutSAM-Eval | ImageReward: 60.88 -> 58.34 | Hurt |
| drop detection score from layout reward (ablation) | GRPO HPS+VQA · LayoutSAM-Eval | Precision: 81.84 -> 74.50 | Hurt |
| drop VQA score from layout reward (ablation) | GRPO HPS+DET · LayoutSAM-Eval | Color: 87.35 -> 85.61 | Hurt |
| allow layout tokens to attend to global prompt (vs w/o LP) (ablation) | structured masking · LayoutSAM-Eval | attribute control | Improved |
| use per-object causal, isolated layout masks and region-wise image attention (vs w/o Local Causal) (ablation) | structured masking · LayoutSAM-Eval | spatial metrics | Improved |
| SMARLI vs PlanGen | AR layout-to-image · OverLayBench-Complex | SR_E: 67.79 -> 80.52 | Improved |
| SMARLI vs PlanGen on FID | AR layout-to-image · LayoutSAM-Eval | FID: 13.91 -> 18.16 | Hurt |
| add layout condition on top of Show-o | Show-o · T2I-CompBench | Spatial: 39.6 -> 49.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PlanGen (1.5B) — PlanGen: Towards Unified Layout Planning and Image Generation in Auto-Regressive Vision Language Models (ICCV 2025) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention (NeurIPS 2025 · similarity 0.88)
- OverLayBench: A Benchmark for Layout-to-Image Generation with Dense Overlaps (NeurIPS 2025 · similarity 0.88)
- Hawk: Leveraging Spatial Context for Faster Autoregressive Text-to-Image Generation (NeurIPS 2025 · similarity 0.87)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.87)
- Control and Realism: Best of Both Worlds in Layout-to-Image without Training (ICML 2025 · similarity 0.87)
- SpotActor: Training-Free Layout-Controlled Consistent Image Generation (AAAI 2025 · 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.
