arXiv cs.AI· Wenwu Tang, Dong Wang, Yun Cheng, Olga Saukh·· 13 小时前AI 评分31
ReSeed：仅调第一步即可恢复扩散模型蒸馏丢失的多样性
A First-Step Bias Restores Lost Diversity in Diffusion Model Distillation
AI 导读
研究者提出 ReSeed，一种后蒸馏方法，只在学生模型第一步去噪上添加 seed-response 偏置适配器，并对齐教师模型首个区间的割线速度，其余权重与后续步骤全部冻结。
正文 · 原文
Abstract:Distribution-matching distillation enables high-quality image generation in only a few denoising steps. However, distilled student models often map different initial noise samples to similar outputs under the same prompt. To locate where diversity is lost, we compare the teacher and student along the sampling trajectory. The largest measured reduction in the response to different initial noises occurs during the first, high-noise step. When we replace the first step with the teacher's corresponding trajectory and leave the final three student steps unchanged, most of the output diversity returns. This suggests that the first step is a key point for restoring diversity. Motivated by this finding, we introduce \emph{ReSeed}, a post-distillation method that adds a seed-response bias adapter only to the first student step and aligns its velocity with the teacher's first-interval secant velocity. All student weights and later denoising steps remain unchanged. We evaluate ReSeed on three text-to-image models (SD3-Medium, SD3.5-Medium, and FLUX.1-dev) and one text-to-video model (Wan2.1-T2V-1.3B). For a distilled 4-step FLUX.1-dev model, \emph{ReSeed} requires only $0.00196\%$ additional trainable parameters relative to the base model and is trained on 8,000 prompt--noise pairs, while keeping the distilled student fully frozen and requiring no target images, perceptual losses, or adversarial losses. It improves DINO diversity by $48.3\%$ while maintaining competitive visual quality. These results show that substantial diversity can be recovered through a lightweight, data-efficient tuning of only the first denoising step, without finetuning the full student model.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.37110 [cs.CV] |
| (or arXiv:2610.37110v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.37110 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wenwu Tang [view email]
[v1]
Tue, 27 Oct 2026 12:08:03 UTC (27,082 KB)
来源：arXiv cs.AI · arxiv.org