Research map / Image generation / Text-to-image diffusion
Penalizing Boundary Activation for Object Completeness in Diffusion Models
Haoyang Xu, Tianhao Zhao, Sibei Yang, Yutian Lin
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 cross-attention constraint (ablation) | boundary activation penalty on SD2.1 · HOIR | HOIR: 38.7 % -> 17.3 % | Improved |
| add self-attention constraint (ablation) | boundary activation penalty on SD2.1 · HOIR | HOIR: 34.7 % -> 17.3 % | Improved |
| add cross-attention constraint (ablation) | boundary activation penalty · CLIP-IQA | CLIP-IQA: 0.643 -> 0.700 | Improved |
| add self-attention constraint (ablation) | boundary activation penalty · CLIP-IQA | CLIP-IQA: 0.681 -> 0.700 | Improved |
| apply penalizing boundary activation | SDXL · Incompleteness 1,200 prompts | HOIR: 18.3 % -> 7.1 % | Improved |
| apply penalizing boundary activation | SD1.5 · Incompleteness 1,200 prompts | HOIR: 43.7 % -> 18.7 % | Improved |
| apply penalizing boundary activation | SD2.1 · Object completeness | HOIR: 45.7 % -> 17.3 % | Improved |
| apply penalizing boundary activation | SD2.1 · Image quality | CLIP-IQA: 0.714 -> 0.703 | Hurt |
| apply penalizing boundary activation | SD2.1 · Time cost | Time cost (s): 5.51 -> 5.75 | Hurt |
| add completeness adjectives to prompt | SD2.1 · Prompt type | HOIR: 45.7 % -> 43.2 % | No clear effect |
| add 'in the middle' to prompt | SD2.1 · Prompt type | HOIR: 45.7 % -> 44.3 % | No clear effect |
| fine-tune with RandomCrop augmentation versus original images (ablation) | SD2.1 fine-tuned on 10,000 LAION images · Incompleteness rate | HOIR | Hurt |
Closest papers by meaning
- The Crystal Ball Hypothesis in diffusion models: Anticipating object positions from initial noise (ICLR 2025 · similarity 0.90)
- Correcting Diffusion Generation through Resampling (CVPR 2024 · similarity 0.88)
- Smooth Diffusion: Crafting Smooth Latent Spaces in Diffusion Models (CVPR 2024 · similarity 0.88)
- Fast constrained sampling in pre-trained diffusion models (NeurIPS 2025 · similarity 0.88)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.87)
- On Error Propagation of Diffusion Models (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.
