通过正向激励噪声提升生成模型性能，仅增少量参数即显著优化结果。
Rectified Noise: A Generative Model Using Positive-incentive Noise
- 在预训练流模型中注入正向激励噪声，改进生成过程。
- ImageNet-1k上FID从10.16降至9.05，生成质量明显提升。
- 新增参数仅0.39%，适合快速部署到现有生成模型中。
Rectified Flow（RF）作为一种高效的生成模型被广泛应用。尽管RF主要基于概率流常微分方程（ODE），但近期研究表明，通过反向时间随机微分方程（SDE）注入噪声进行采样可实现更优的生成效果。受正向激励噪声（pi-noise）启发，我们提出一种新型生成算法——矩形噪声（Rectified Noise, RN），用于训练pi-noise生成器，通过将pi-noise注入预训练RF模型的速度场来提升生成性能。引入矩形噪声流程后，预训练的RF模型可高效转化为pi-noise生成器。我们在多种模型架构和数据集上进行了广泛实验验证。结果表明：（1）使用矩形噪声的RF模型在ImageNet-1k上的FID从10.16降至9.05；（2）pi-noise生成器仅增加0.39%的训练参数，即可实现性能提升。
原文摘要 · Abstract (English)
Rectified Flow (RF) has been widely used as an effective generative model. Although RF is primarily based on probability flow Ordinary Differential Equations (ODE), recent studies have shown that injecting noise through reverse-time Stochastic Differential Equations (SDE) for sampling can achieve superior generative performance. Inspired by Positive-incentive Noise (pi-noise), we propose an innovative generative algorithm to train pi-noise generators, namely Rectified Noise (RN), which improves the generative performance by injecting pi-noise into the velocity field of pre-trained RF models. After introducing the Rectified Noise pipeline, pre-trained RF models can be efficiently transformed into pi-noise generators. We validate Rectified Noise by conducting extensive experiments across various model architectures on different datasets. Notably, we find that: (1) RF models using Rectified Noise reduce FID from 10.16 to 9.05 on ImageNet-1k. (2) The models of pi-noise generators achieve improved performance with only 0.39% additional training parameters.
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。