For the domain adaptation I use the Deep Reconstruction-Classification Network (DRCN).
The model is based on a convolutional architecture that has two pipelines with a shared encoding representation. First pipeline is a convolutional network for label prediction based on the source data, second pipeline is a convolutional autoencoder for target data reconstruction. Including the reconstruction of target data among with a standard label classifier helps to implement the domain adaptation. The model is based on a paper and a code.
