Domain Adaptation and Refinement for Synthetic Data
Domain adaptation (DA) is a set of techniques designed to make a model trained on one domain of data, the source domain, work well on a different, target domain. This is a natural fit for synthetic data: in almost all applications, we would like to train the model in the source domain of synthetic data but then apply the results in the target domain of real data. In this talk, we give a survey of DA approaches that have been used for such synthetic-to-real adaptation, concentrating on deep learning models. We will see the gaze estimation story from Apple’s Refiner and beyond, DA techniques for learning to drive, GAN-based DA for medical imaging, and much more. Expect a lot of GANs and loss functions!