Research map / Image generation / Fast diffusion sampling
Spectral Evolution Search: Efficient Inference-Time Scaling for Reward-Aligned Image Generation
Jinyan Ye, Zhongjie Duan, Zhiwen Li, Cen Chen, Daoyuan Chen, Yaliang Li, Yingda Chen
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 |
|---|---|---|---|
| search low-frequency wavelet subspace with CEM instead of Best-of-N | SDXL · DrawBench Aesthetic, NRE=200 | Aesthetic: 6.35 -> 6.55 | Improved |
| search low-frequency wavelet subspace with CEM instead of Best-of-N | FLUX.1-dev · Pick-a-Pic Aesthetic, NRE=200 | Aesthetic: 6.80 -> 7.03 | Improved |
| search low-frequency wavelet subspace with CEM instead of Best-of-N | SDXL · DrawBench CLIP, NRE=200 | CLIP score: 43.30 -> 43.53 | Improved |
| restrict search to LL subband instead of high-frequency subbands (ablation) | SES · SDXL Aesthetic ablation | Aesthetic: -0.65 | Improved |
| restrict search to LL subband instead of full frequency space (ablation) | SES · SDXL Aesthetic ablation | Aesthetic: -0.53 | Improved |
| use LL subband instead of random subspace of equal dimension (ablation) | SES · SDXL Aesthetic ablation | Aesthetic: -0.83 | Improved |
| use decomposition level J=4 instead of J=2 (ablation) | SES · SDXL Aesthetic ablation | Aesthetic: -0.11 | Improved |
| use decomposition level J=4 instead of J=6 (ablation) | SES · SDXL Aesthetic ablation | Aesthetic: -0.20 | Improved |
| use CEM instead of random search (ablation) | SES · SDXL Aesthetic ablation | Aesthetic: -0.21 | Improved |
| apply SES to DPO-finetuned model | SDXL + DPO · Aesthetic | Aesthetic: 5.59 -> 6.40 | Improved |
| apply SES to SPO-finetuned model | SDXL + SPO · Aesthetic | Aesthetic: 5.90 -> 6.78 | Improved |
| increase budget from NRE=200 to 1000 with SES versus BoN | SES / BoN · SDXL Aesthetic scaling | Aesthetic | Mixed |
Closest papers by meaning
- SeaCache: Spectral-Evolution-Aware Cache for Accelerating Diffusion Models (CVPR 2026 · similarity 0.89)
- Equilibrium Matching: Generative Modeling with Implicit Energy-Based Models (NeurIPS 2026 · similarity 0.86)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.86)
- TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generation (NeurIPS 2025 · similarity 0.86)
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.86)
- Adaptive Inference-Time Scaling via Cyclic Diffusion Search (NeurIPS 2025 · 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.
