Velocity scaling improves image generation quality by correcting time lag
Velocity Scaling in Flow Matching
Computer Vision and Pattern Recognition
Summary
Creating images using a model that simulates changing states over time can be tricky because the model's imagined progress can lag behind real training progress. The authors show that adjusting how fast the changes happen helps fix this lag, making generated images much better. They found that the common idea that the model underestimates change size is incorrect. Instead, changing the speed helps the model match the right stage of image creation, leading to clearer pictures.
What this means in practice
- •For deep learning engineers: Improve image synthesis by adjusting velocity scaling to reduce time lag in flow matching models.
- •For computer vision teams: Enhance generative image quality in production by selecting velocity gain based on population time lag measurements.
Authors
Youssef Saied, François Fleuret
Abstract
Scaling a learned flow-matching velocity field $v_θ$ by a gain $γ(t)$ was recently shown to greatly improve generation quality. Prior work argued that velocity fields trained with mean-squared error (MSE) systematically underestimate velocity magnitude and that scaling corrects this error. We show that MSE training does not create a velocity-magnitude deficit. We find instead that velocity scaling reduces population time lag: sampled states at model time $t$ resemble training states from an earlier time. Velocity scaling and moving model time back are two ways to address this population time lag. Across architectures and model sizes, measuring population time lag and using it to select a gain greatly improves generation quality, reducing FID from 28.0 to 12.2 (estimated by linear interpolation between FID measurements at neighboring gains) on ImageNet-256 at NFE 25 without guidance.