改进扩散模型采样误差，提升图像生成质量
CFG-OEC: Classifier Free Guidance with Orthogonal Error Correction
- 通过正交误差修正降低条件与无条件预测的干扰
- 在多个采样器和引导强度下优于CFG和CFG++
- 适合追求高质量图像生成的研究者与开发者
Classifier free guidance（CFG）是扩散模型中条件采样的标准方法，但其采样规则与训练目标不一致，导致条件与无条件预测误差交互引发结构性采样误差。本文将采样误差分解为基项与交叉项，交叉项由两类误差对齐程度决定。基于此分析，提出CFG-OEC，通过结构化修改减少交叉项。针对真实噪声不可观测的实际情况，引入基于模型预测的代理量及动态校正方法，稳定跨扩散步数的修正效果。控制环境下实验验证了理论误差分解与代理构造的有效性。在Stable Diffusion v1.5和Stable Diffusion XL上的图像生成任务中，CFG-OEC在多种采样器和引导强度下均显著优于CFG和CFG++，FID与CLIP分数更优。
原文摘要 · Abstract (English)
Classifier free guidance is a standard method for conditional sampling in diffusion models, but its sampling rule is not aligned with the objective used in training. This mismatch induces a structural sampling error through the interaction of conditional and unconditional prediction errors. We analyze this issue by decomposing the sampling error into a base term and a cross term determined by the alignment of the two errors. Based on this analysis we propose CFG with orthogonal error correction (CFG-OEC), a structural modification that reduces the interaction term. For practical settings where ground truth noise is not observable, we introduce a proxy computed from model predictions and a dynamic method that stabilizes correction across diffusion timesteps. Experiments in a controlled environment validate our theoretical error decomposition and proxy construction. Image generation on Stable Diffusion v1.5 and Stable Diffusion XL show that CFG-OEC improves FID and CLIP scores over CFG and CFG++ across multiple samplers and guidance regimes.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。