Algorithm 1 MixDANN for Domain Generalization

$$\begin{array}{cc}\hfill & \mathrm{Data}:\text{Set of sample}x\text{and label}y\text{from Sources}{\mathbb{P}}_1,\dots ,{\mathbb{P}}_k\text{and Target}\mathbb{Q}\hfill \\ \hfill & \mathrm{Models}:\text{U-Net (Feature Extractor}{r}_{\theta }\text{and Task-specific Network}{c}_{\sigma }),\text{Do-}\hfill \\ \hfill & \text{main Discriminator}{d}_{\mu }\hfill \\ \hfill & 1:\mathbf{procedure}\text{TRAIN MIXDANN USING SOURCES}({\mathbb{P}}_i{)}_i\hfill \\ \hfill & 2:\mathbf{for}\text{mini-batch}\text{of}\text{samples from}\text{sources}(x_i{)}_i\sim ({\mathbb{P}}_i{)}_i\mathbf{do}\hfill \\ \hfill & 3:\mathrm{Mixup}:{\tilde{x}}_{p,q}\leftarrow \lambda x_p+(1-\lambda )x_q\text{for}p,q\text{pairs},\lambda \sim Beta(0.7,0.7)\hfill \\ \hfill & 4:\mathbf{Task}\mathrm{Update}:\theta ,\sigma \leftarrow {\text{arg min}}_{\sigma ,\theta }\sum {\mathcal{L}}_T(\sigma ,\theta ,{\tilde{x}}_{i,j})\hfill \\ \hfill & 5:\mathbf{Domain}\mathrm{Update}:\theta ,\mu \leftarrow \text{arg}[{\text{min}}_{\theta }{\max }_{\mu }-\gamma \sum {\mathcal{L}}_D(\mu ,\theta ,{\tilde{x}}_{i,j})]\hfill \\ \hfill & 6:\mathbf{end}\mathbf{for}\hfill \\ \hfill & 7:\mathbf{end}\mathbf{procedure}\hfill \\ \hfill & 8:\mathbf{procedure}\text{TESTING MIXDANN ON TARGET}\mathbb{Q}\hfill \\ \hfill & 9:\mathrm{Prediction}:y_i\leftarrow c_{\sigma }\circ r_{\theta }(x_i)\text{for samples from target}{x}_i\sim \mathbb{Q}\hfill \\ \hfill & 10:\mathbf{end}\mathbf{procedure}\hfill \end{array}$$