Unsupervised Pixel-Level Domain Adaptation with Generative Adversarial Networks

Collecting well-annotated image datasets to train modern machine learning is prohibitively expensive for many tasks. One appealing alternative rendering synthetic data where ground-truth annotations are generated. Unfortunately, models trained purely on rendered images often to generalize to real images. To address this shortcoming, prior work unsupervised domain adaptation algorithms that attempt to map between the two domains or learn to extract features that are-invariant. In this work, we present a new approach that learns, in an manner, a transformation in the pixel space from one domain to the. Our generative adversarial network (GAN)-based method adapts-domain images to appear as if drawn from the target domain. Our approach only produces plausible samples, but also outperforms the state-of-the-art a number of unsupervised domain adaptation scenarios by large margins., we demonstrate that the adaptation process generalizes to object unseen during training.