Research map / Image generation / Text-to-image diffusion
I-DRUID: Layout to image generation via instance-disentangled representation and unpaired data
Fengxiang Yang, Tianyi Zheng, Bangjie Yin, Shice Liu, Jinwei Chen, Peng-Tao Jiang, Bo Li
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 IDM with instance disentanglement constraint (ablation) | SD3 L2I baseline · COCO-MIG | Avg ISR: 56.82 -> 57.64 | Improved |
| add IDM with instance disentanglement constraint (ablation) | SD3 L2I baseline · LayoutSAM-eval | Spatial: 86.96 -> 88.53 | Improved |
| add IDM with instance disentanglement constraint (ablation) | SD3 L2I baseline · LayoutSAM-eval | Shape: 72.44 -> 75.62 | Improved |
| add PPO-based RL on unpaired prompts to IDM model (ablation) | I-DRUID SD3 · COCO-MIG | Avg ISR: 57.64 -> 62.75 | Improved |
| add PPO-based RL on unpaired prompts to IDM model (ablation) | I-DRUID SD3 · LayoutSAM-eval | Shape: 75.62 -> 77.20 | Improved |
| add IDM to PPO-based RL (ablation) | RL-PPO · COCO-MIG | Avg ISR: 60.23 -> 62.75 | Improved |
| add PPO RL without IDM to baseline (ablation) | SD3 L2I baseline · COCO-MIG | Avg ISR: 56.82 -> 60.23 | Improved |
| use SFT on paired COCO-MIG data instead of unpaired RL (ablation) | IDM model · COCO-MIG | Avg ISR: 66.92 vs 62.75 | Mixed |
| use SFT on paired COCO-MIG data instead of unpaired RL (ablation) | IDM model · LayoutSAM-eval | Texture: 74.33 vs 78.35 | Hurt |
| replace PPO with GRPO (ablation) | IDM+RL · COCO-MIG | Avg ISR: 61.64 vs 62.75 | No clear effect |
| replace PPO with GRPO (ablation) | IDM+RL · training speed | seconds per iteration: 15s/iter -> more than 60s/iter | Hurt |
| lower lambda_dis to 0.5 with lambda_rl 1 (ablation) | I-DRUID · COCO-MIG | Avg ISR: performance degradation than baseline | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MIGC — MIGC: Multi-Instance Generation Controller for Text-to-Image Synthesis (CVPR 2024) · better on 3 of 3 shared results
- HiCo — HiCo: Hierarchical Controllable Diffusion Model for Layout-to-image Generation (NeurIPS 2024) · better on 1 of 1 shared result
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
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.89)
- Disentangled 3D Scene Generation with Layout Learning (ICML 2024 · similarity 0.88)
- InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention (NeurIPS 2025 · similarity 0.88)
- CoT-lized Diffusion: Let's Reinforce T2I Generation Step-by-step (NeurIPS 2025 · similarity 0.87)
- OverLayBench: A Benchmark for Layout-to-Image Generation with Dense Overlaps (NeurIPS 2025 · similarity 0.86)
- Adversarial Supervision Makes Layout-to-Image Diffusion Models Thrive (ICLR 2024 · 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.
