[论文] Normalizing Trajectory Models

## 论文概要 **研究领域**: CV **作者**: Jiatao Gu, Tianrong Chen, Ying Shen **发布时间**: 2025-05-07 **arXiv**: [2505.05129](https://arxiv.org/abs/2505.05129) ## 中文摘要 基于扩散的模型将采样分解为多个小的高斯去噪步骤——这一假设在生成被压缩为少量粗粒度转换时失效。现有的少步方法通过蒸馏、一致性训练或对抗目标来解决这个问题，但在此过程中牺牲了似然框架。我们引入归一化轨迹模型（NTM），将每个反向步骤建模为富有表现力的条件归一化流，并进行精确的似然训练。在架构上，NTM将浅层可逆块与深度并行预测器结合在轨迹上，形成端到端网络，可从零开始训练或从预训练的流匹配模型初始化。其精确的轨迹似然进一步支持自蒸馏：在模型自身得分上训练的轻量级去噪器在四步内产生高质量样本。在文本到图像基准上，NTM在仅四步采样内达到或超过强图像生成基线，同时独特地保留了生成轨迹上的精确似然。 ## 原文摘要 Diffusion-based models decompose sampling into many small Gaussian denoising steps -- an assumption that breaks down when generation is compressed to a few coarse transitions. Existing few-step methods address this through distillation, consistency training, or adversarial objectives, but sacrifice the likelihood framework in the process. We introduce Normalizing Trajectory Models (NTM), which models each reverse step as an expressive conditional normalizing flow with exact likelihood training. Architecturally, NTM combines shallow invertible blocks within each step with a deep parallel predictor across the trajectory, forming an end-to-end network trainable from scratch or initializable from pretrained flow-matching models. Its exact trajectory likelihood further enables self-distillation... --- *自动采集于 2026-05-12* #论文 #arXiv #CV #小凯