While deep learning techniques are nowadays commonly used in the field of computer vision and have achieved outstanding performance, collecting sufficient data for training a deep neural network remains a challenge. In this work, domain adaptation techniques are utilized in order to learn from a different yet related domain, where data are more readily available. In this thesis, a framework called Synergistic Image and Feature Adaptation (SIFA) is implemented using PyTorch, and in the project it is demonstrated how to perform semantic segmentation task on light microscopy images of fungal-colonized root sections. This helps the analysis of arbuscular mycorrhizal fungi (AMF). The training dataset is composed of images from two domains: unlabeled real-world light microscopy images and labeled synthetic images. The synthetic images and the labels are acquired from a 3D model of a root section. A baseline model adopting the U-Net architecture trained with the synthetic images suffers from severe performance degradation when applied to the light microscopy images, while the segmentation model developed in this work is able to create better segmentations, and the result is overall promising.