扩散采样路径透露更多：一种高效的即插即用样本筛选策略
Diffusion Sampling Path Tells More: An Efficient Plug-and-Play Strategy for Sample Filtering
提出用 CFG 过程中条件与无条件分数的累积差异（ASD）作为样本质量的无监督代理指标，可在去噪早期直接拒绝低质量样本，无需外部奖励模型或重训练；做文生图采样加速与质量筛选时可直接嵌入现有采样循环。
原文依据 we identify a strong correlation between high-density regions of the sample distribution and the Accumulated Score Differences (ASD)
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论扩散模型常因采样轨迹中固有的随机变化而表现出样本质量不稳定。尽管基于训练的微调和推理时对齐技术旨在提升样本保真度，但它们通常需要完整的去噪过程和外部奖励信号，这带来可观的计算开销，阻碍了其更广泛的应用。在本工作中，我们揭示了一个有趣的现象：在无分类器引导（CFG）过程中，样本质量与去噪轨迹特征之间存在此前未被观察到却可利用的联系。具体而言，我们发现样本分布的高密度区域与累积分数差异（ASD）——即条件分数与无条件分数之间的累积散度——之间存在强相关性。基于这一洞见，我们提出 CFG-Rejection，一种高效、即插即用的策略，可在去噪过程的早期阶段筛选出低质量样本，且关键是不需要外部奖励信号或模型重训练。重要的是，我们的方法无需修改模型架构或采样调度，并与现有扩散框架保持完全兼容。我们通过大量实验验证了 CFG-Rejection 在图像生成中的有效性，在人类偏好分数（HPSv2、PickScore）和具有挑战性的基准（GenEval、DPG-Bench）上均取得显著提升。我们预期 CFG-Rejection 将为图像之外的多种生成模态带来显著优势，为更高效、更可靠的高质量样本生成铺平道路。
英文原文摘要 1658 字符
Diffusion models often exhibit inconsistent sample quality due to stochastic variations inherent in their sampling trajectories. Although training-based fine-tuning and inference-time alignment techniques aim to improve sample fidelity, they typically necessitate full denoising processes and external reward signals. This incurs substantial computational costs, hindering their broader applicability. In this work, we unveil an intriguing phenomenon: a previously unobserved yet exploitable link between sample quality and characteristics of the denoising trajectory during classifier-free guidance (CFG). Specifically, we identify a strong correlation between high-density regions of the sample distribution and the Accumulated Score Differences (ASD)--the cumulative divergence between conditional and unconditional scores. Leveraging this insight, we introduce CFG-Rejection, an efficient, plug-and-play strategy that filters low-quality samples at an early stage of the denoising process, crucially without requiring external reward signals or model retraining. Importantly, our approach necessitates no modifications to model architectures or sampling schedules and maintains full compatibility with existing diffusion frameworks. We validate the effectiveness of CFG-Rejection in image generation through extensive experiments, demonstrating marked improvements on human preference scores (HPSv2, PickScore) and challenging benchmarks (GenEval, DPG-Bench). We anticipate that CFG-Rejection will offer significant advantages for diverse generative modalities beyond images, paving the way for more efficient and reliable high-quality sample generation.