This thesis presents research conducted in the area of synthetic data generation for computer vision tasks. The research aims to address the challenge of datahungry deep learning models by generating synthetic images that can effectively train computer vision models to solve tasks such as object counting, polyp segmentation, and pattern classification. The work carried out explores the use of various techniques to ensure effective use of synthetic data, including domain randomisation and domain adaptation in both self- and semi-supervised setups. Through the application of these techniques, the research aims to develop a robust and effective approach for generating synthetic data that can improve the performance of computer vision models with a reduced amount of human annotations.