Generative Vision Atlas

Objective option

Flow matching / rectified flow

Directly regress the velocity field of a (near-)straight path between a noise sample and a data sample, rather than reversing a stochastic noising process.

Introduced by

Used by (5)

Flow Matching, RAE, Rectified Flow, Scale-RAE, SD3 / MMDiT

Alternatives on this axis

Diffusion-loss autoregression, Diffusion (epsilon/v/x0-prediction), Drifting field, Stochastic interpolant framework, Average (mean) velocity, Next-scale autoregression, Normalizing flow, Transition matching

← All Objective options