Elucidating the SNR-t Bias of Diffusion Probabilistic Models
This work addresses a fundamental bias in diffusion models that degrades generation quality, offering a simple fix applicable to many existing models.
Diffusion models suffer from a Signal-to-Noise Ratio-timestep (SNR-t) bias during inference, causing error accumulation and degraded generation quality. The authors propose a differential correction method that decomposes samples into frequency components and applies correction individually, improving generation quality across multiple diffusion models (IDDPM, ADM, DDIM, etc.) with negligible computational overhead.
Diffusion Probabilistic Models have demonstrated remarkable performance across a wide range of generative tasks. However, we have observed that these models often suffer from a Signal-to-Noise Ratio-timestep (SNR-t) bias. This bias refers to the misalignment between the SNR of the denoising sample and its corresponding timestep during the inference phase. Specifically, during training, the SNR of a sample is strictly coupled with its timestep. However, this correspondence is disrupted during inference, leading to error accumulation and impairing the generation quality. We provide comprehensive empirical evidence and theoretical analysis to substantiate this phenomenon and propose a simple yet effective differential correction method to mitigate the SNR-t bias. Recognizing that diffusion models typically reconstruct low-frequency components before focusing on high-frequency details during the reverse denoising process, we decompose samples into various frequency components and apply differential correction to each component individually. Extensive experiments show that our approach significantly improves the generation quality of various diffusion models (IDDPM, ADM, DDIM, A-DPM, EA-DPM, EDM, PFGM++, and FLUX) on datasets of various resolutions with negligible computational overhead. The code is at https://github.com/AMAP-ML/DCW.