On the Importance of Noise Scheduling for Diffusion Models
This work addresses the problem of optimizing noise scheduling for high-resolution image generation in diffusion models, offering a simple recipe that improves performance, though it is incremental as it builds on existing methods.
The study found that noise scheduling is crucial for diffusion model performance, with optimal schedules depending on task specifics like image size, and that scaling input data while keeping the noise schedule fixed is effective across sizes, leading to state-of-the-art results with 1024x1024 image generation on ImageNet.
We empirically study the effect of noise scheduling strategies for denoising diffusion generative models. There are three findings: (1) the noise scheduling is crucial for the performance, and the optimal one depends on the task (e.g., image sizes), (2) when increasing the image size, the optimal noise scheduling shifts towards a noisier one (due to increased redundancy in pixels), and (3) simply scaling the input data by a factor of $b$ while keeping the noise schedule function fixed (equivalent to shifting the logSNR by $\log b$) is a good strategy across image sizes. This simple recipe, when combined with recently proposed Recurrent Interface Network (RIN), yields state-of-the-art pixel-based diffusion models for high-resolution images on ImageNet, enabling single-stage, end-to-end generation of diverse and high-fidelity images at 1024$\times$1024 resolution (without upsampling/cascades).