去掉扩散模型的时间嵌入，仍能生成高质量图像。
On the Redundancy of Timestep Embeddings in Diffusion Models
- 通过理论分析与实验证明，时间嵌入非必需。
- 在CelebA和CIFAR-10上，去时模型FID等指标更优。
- 适合追求高效生成架构的研究者参考。
扩散模型依赖显式的时间步嵌入来调节不同噪声尺度下的去噪过程。本文挑战这一设计的必要性，分析其对U-Net和Diffusion Transformer架构的影响。除实证证据外，我们提供理论框架，证明在特定条件下，扩散训练目标的全局最小值可不依赖显式时间条件达成。消融实验显示，在CelebA和CIFAR-10数据集上，完全移除时间嵌入后，模型仍保持高结构保真度，且在FID、精确率和召回率等指标上甚至优于带时间条件的基线。分析表明，在特定假设下，这些架构可从含噪输入中隐式推断噪声尺度，使显式时间条件冗余。本研究颠覆了长期依赖时间条件的范式，为更高效、聚焦结构的生成架构开辟路径。
原文摘要 · Abstract (English)
Diffusion models rely heavily on explicit timestep embeddings to modulate the denoising process across various noise scales. In this work, we challenge the necessity of these temporal signals by analyzing their impact on U-Net and Diffusion Transformer architectures. Beyond empirical evidence, we provide a theoretical framework demonstrating that, under certain conditions, the global minimizer of the diffusion training objective can be achieved without explicit timestep conditioning. Our findings reveal a surprising robustness when timestep embeddings are completely removed. Extensive ablation studies on the CelebA and CIFAR-10 datasets show that these time-agnostic models can maintain high structural fidelity and even surpass their conditioned counterparts in competitive metrics, including FID, precision, and recall. Our analysis suggests these architectures can implicitly infer noise scales from the corrupted input under specific assumptions, rendering explicit temporal conditioning redundant. This study challenges long-standing temporal conditioning paradigms and paves the way for more efficient and structurally focused generative architectures.
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。