Fig. 7
From: Generative super-resolution of turbulent flows via stochastic interpolants
Convergence of the Wasserstein-1 distance of the density of \(\langle \epsilon \rangle\) as a function of the number of pseudo-timesteps used to infer super-resolutions. The SI-models are observed to converge for a lower number of pseudo-timesteps compared to their contesting counterparts. For the FM- and DM-models, the \(W_1\)-value \(10^3\) is used as a placeholder, as the models failed to produce meaningful solutions when limited to only 10 timesteps during the inference stage.
