Research map / Image generation / Text-to-image diffusion
The Lottery Ticket Hypothesis in Denoising: Towards Semantic-Driven Initialization
Jiafeng Mao, Xueting Wang, Kiyoharu Aizawa
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 |
|---|---|---|---|
| construct initial noise from winning tickets (SemI) instead of random noise | Stable Diffusion v1.4 · MS COCO 2017 layout control | R_suc (%): 12.14 -> 21.94 | Improved |
| combine SemI initial noise with Soft mask guidance | Soft · MS COCO 2017 layout control | R_suc (%): 22.31 -> 28.73 | Improved |
| combine SemI initial noise with Swap guidance | Swap · MS COCO 2017 layout control | R_suc (%): 21.58 -> 23.73 | Improved |
| increase number of initial noise images N from 10 to 100 (ablation) | SemI · MS COCO 2017 layout control | R_suc (%): 21.76 -> 21.94 | Improved |
| raise object threshold T_obj from 0.5 to 0.7 (ablation) | SemI · MS COCO 2017 layout control (large objects) | R_suc^l (%): 52.31 -> 47.59 | Hurt |
| raise object threshold T_obj from 0.5 to 0.7 (ablation) | SemI · MS COCO 2017 layout control (small objects) | R_suc^s (%): 2.40 -> 3.05 | Improved |
| lower background threshold T_bg from 0.1 to 0.05 (ablation) | SemI · MS COCO 2017 layout control (small objects) | R_suc^s (%): 2.40 -> 3.02 | Improved |
| use winning tickets versus Soft on large objects | SemI · MS COCO 2017 layout control (large objects) | R_suc^l (%): 56.24 -> 52.31 (Soft vs SemI) | Hurt |
Closest papers by meaning
- InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR 2024 · similarity 0.89)
- The Crystal Ball Hypothesis in diffusion models: Anticipating object positions from initial noise (ICLR 2025 · similarity 0.88)
- Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Rethinking Training for De-biasing Text-to-Image Generation: Unlocking the Potential of Stable Diffusion (CVPR 2025 · similarity 0.87)
- Towards Understanding the Working Mechanism of Text-to-Image Diffusion Model (NeurIPS 2024 · similarity 0.86)
- Cyclic Denoising Reveals Ultrastable Memories in Diffusion Models (NeurIPS 2026 · 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.
