Algorithm 1. Seg-JDOT stochastic optimization

Require: xs: source domain images, xt: target domain images, ys: source domain segmentation maps

for each source batch $({\text{x}}_b^s,{\text{y}}_b^s)$ and target batch (${\text{x}}_b^s$) do

fix $\widehat{g}$ and $\widehat{f}$, find γ for the given batch

fix $\widehat{\gamma }$, and use gradient descent to update $\widehat{f}$ and $\widehat{g}$

end for