Research map / Image generation / Text-to-image diffusion
Mitigating Noise-Induced Layout Priors for Object Counting in Diffusion Models
Xiaoling Gu, Xuelong Li, Shengqi Wu, Yongkang Wong, Zizhao Wu, Huan Li, Zhou Yu, Mohan Kankanhalli
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 Object Saliency Enhancement (ablation) | Ours (SDXL) · T2I-Count | accuracy: 0.5378 -> 0.5556 | Improved |
| add LRR and OSE (count-aware noise adjustment) (ablation) | Ours (SDXL) · T2I-Count | accuracy: 0.5022 -> 0.5556 | Improved |
| add Attention Concentration Loss (ablation) | Ours (SDXL) · T2I-Count | accuracy: 0.3622 -> 0.5556 | Improved |
| add Dispersion Regularization Loss (ablation) | Ours (SDXL) · T2I-Count | accuracy: 0.5311 -> 0.5556 | Improved |
| add both attention losses (layout consistency strategy) (ablation) | Ours (SDXL) · T2I-Count | accuracy: 0.3511 -> 0.5556 | Improved |
| add Attention Concentration Loss (ablation) | Ours (SDXL) · T2I-Count | STA: 0.6701 -> 0.8123 | Improved |
| add Object Saliency Enhancement (ablation) | Ours (SDXL) · T2I-Count | MAE: 1.1267 -> 0.9889 | Improved |
| apply the proposed training-free counting framework | SDXL · COCOCount | accuracy: 0.2958 -> 0.6125 | Improved |
| apply the proposed training-free counting framework | SD v3 · COCOCount | accuracy: 0.5058 -> 0.6225 | Improved |
| apply the proposed training-free counting framework | SDXL · T2I-Count | accuracy: 0.2978 -> 0.5556 | Improved |
| apply the proposed training-free counting framework | SD v3 · T2I-Count | accuracy: 0.4444 -> 0.6022 | Improved |
| apply the proposed training-free counting framework | SD v3 · MultiCount | accuracy: 0.1486 -> 0.3500 | Improved |
Closest papers by meaning
- The Crystal Ball Hypothesis in diffusion models: Anticipating object positions from initial noise (ICLR 2025 · similarity 0.89)
- Make It Count: Text-to-Image Generation with an Accurate Number of Objects (CVPR 2025 · similarity 0.89)
- NoiseCollage: A Layout-Aware Text-to-Image Diffusion Model Based on Noise Cropping and Merging (CVPR 2024 · similarity 0.89)
- InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR 2024 · similarity 0.88)
- Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Inference-Time Alignment of Diffusion Models with Direct Noise Optimization (ICML 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.
