From Navigation to Refinement: Revealing the Two-Stage Nature of Flow-based Diffusion Models through Oracle Velocity
By: Haoming Liu , Jinnuo Liu , Yanhao Li and more
Potential Business Impact:
Teaches computers to create realistic pictures and videos.
Flow-based diffusion models have emerged as a leading paradigm for training generative models across images and videos. However, their memorization-generalization behavior remains poorly understood. In this work, we revisit the flow matching (FM) objective and study its marginal velocity field, which admits a closed-form expression, allowing exact computation of the oracle FM target. Analyzing this oracle velocity field reveals that flow-based diffusion models inherently formulate a two-stage training target: an early stage guided by a mixture of data modes, and a later stage dominated by the nearest data sample. The two-stage objective leads to distinct learning behaviors: the early navigation stage generalizes across data modes to form global layouts, whereas the later refinement stage increasingly memorizes fine-grained details. Leveraging these insights, we explain the effectiveness of practical techniques such as timestep-shifted schedules, classifier-free guidance intervals, and latent space design choices. Our study deepens the understanding of diffusion model training dynamics and offers principles for guiding future architectural and algorithmic improvements.
Similar Papers
Few-step Flow for 3D Generation via Marginal-Data Transport Distillation
CV and Pattern Recognition
Makes 3D pictures create much faster.
Terminal Velocity Matching
Machine Learning (CS)
Makes AI create amazing pictures super fast.
Terminal Velocity Matching
Machine Learning (CS)
Makes AI create amazing pictures super fast.