提出新训练目标，让扩散模型无需分类器引导即可高效生成高质量图像。
Diffusion Models without Classifier-free Guidance
- 用后验概率建模替代分类器引导，改进扩散模型训练目标。
- 推理速度提升一倍，ImageNet 256 上 FID 达 1.34，超越现有方法。
- 兼容现有模型，易用性强，适合追求高效生成的开发者使用。
本文提出模型引导（Model-guidance, MG），一种用于训练扩散模型的新目标，可消除常用的分类器无关引导（Classifier-free guidance, CFG）。该方法突破了传统仅建模数据分布的范式，引入条件后验概率的建模。其思想源自CFG，但实现更简洁有效，可作为即插即用模块集成到现有模型中。实验表明，该方法显著加速训练过程，推理速度翻倍，并在生成质量上达到卓越表现，优于甚至超过采用CFG的同期扩散模型。在多种模型与数据集上的广泛实验验证了其有效性、效率与可扩展性。最终在ImageNet 256基准上取得最优性能，FID为1.34。代码已开源。
原文摘要 · Abstract (English)
This paper presents Model-guidance (MG), a novel objective for training diffusion model that addresses and removes of the commonly used Classifier-free guidance (CFG). Our innovative approach transcends the standard modeling of solely data distribution to incorporating the posterior probability of conditions. The proposed technique originates from the idea of CFG and is easy yet effective, making it a plug-and-play module for existing models. Our method significantly accelerates the training process, doubles the inference speed, and achieve exceptional quality that parallel and even surpass concurrent diffusion models with CFG. Extensive experiments demonstrate the effectiveness, efficiency, scalability on different models and datasets. Finally, we establish state-of-the-art performance on ImageNet 256 benchmarks with an FID of 1.34. Our code is available at https://github.com/tzco/Diffusion-wo-CFG.
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。