Research map / Image generation / Text-to-image diffusion
Toward Better Out-painting: Improving the Image Composition with Initialization Policy Model
Xuan Han, Yihao Zhao, Yanhao Ge, Mingyu You
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 diffusion reward supervision Lrwd (ablation) | IPM · OpenImage-FCO test (250 cases) | SAM score: SAM score increases by 0.033 when removed (0.103 vs 0.070) | Improved |
| add diffusion reward supervision Lrwd (ablation) | IPM · OpenImage-FCO test (250 cases) | Aesthetic Score: AS dropping by 0.08 when removed (5.26 vs 5.34) | Improved |
| add diffusion reward supervision Lrwd (ablation) | IPM · OpenImage-FCO test (250 cases) | Image Reward: 0.64 -> 0.67 | Improved |
| initialize IPM from pretrained Stable Diffusion U-Net (ablation) | IPM · OpenImage-FCO test (250 cases) | CLIP similarity: CLIP drops by 0.21 when removed (26.29 vs 26.50) | Improved |
| initialize IPM from pretrained Stable Diffusion U-Net (ablation) | IPM · OpenImage-FCO test (250 cases) | SAM score: 0.074 -> 0.070 | No clear effect |
| add IPM to ControlNet baseline (ablation) | ControlNet · OpenImage-FCO test (250 cases) | SAM score: 0.235 -> 0.070 | Improved |
| add IPM to ControlNet baseline (ablation) | ControlNet · OpenImage-FCO test (250 cases) | CLIP similarity: 26.46 -> 26.50 | No clear effect |
| add IPM to ControlNet baseline (ablation) | ControlNet · OpenImage-FCO test (250 cases) | INT (user study): 3.88 -> 4.62 | Improved |
| truncate at ts=700 instead of 800 (ablation) | IPM · OpenImage-FCO test (250 cases) | Image Reward: IR and AS are 0.35 and 0.60 lower than ts = 800 (0.32 vs 0.67) | Hurt |
| truncate at ts=900 instead of 800 (ablation) | IPM · OpenImage-FCO test (250 cases) | SAM score: SAM increases by 0.033 compared to ts = 800 (0.103 vs 0.070) | Hurt |
| truncate at ts=700 instead of 800 (ablation) | IPM · OpenImage-FCO test (250 cases) | SAM score: 0.070 -> 0.047 | Improved |
| fine-tune PowerPaint on OpenImage-FCO | PowerPaint · OpenImage-FCO test (250 cases) | SAM score: 0.820 -> 0.433 | 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.
- BrushNet — BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion (ECCV 2024) · better on 4 of 5 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
- Anywhere: A Multi-Agent Framework for User-Guided, Reliable, and Diverse Foreground-Conditioned Image Generation (AAAI 2025 · similarity 0.82)
- InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR 2024 · similarity 0.82)
- TKG-DM: Training-free Chroma Key Content Generation Diffusion Model (CVPR 2025 · similarity 0.82)
- DRIP: Unleashing Diffusion Priors for Joint Foreground and Alpha Prediction in Image Matting (NeurIPS 2024 · similarity 0.82)
- RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.82)
- Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.82)
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.
