DATA-EFFICIENT LEARNING BY ADVERSARIAL DISCRIMINATIVE DOMAIN ADAPTATION
Technical Report,01 Aug 2017,31 Aug 2019
REGENTS OF THE UNIVERSITY OF CALIFORNIA Berkeley United States
Pagination or Media Count:
The goal of this effort was to streamlined the end-to-end process of using machine learning for AFDoD tasks. Focusing on domain adaptation, University of California at Berkeley developed a discriminatively-trained Cycle-Consistent Adversarial Domain Adaptation model, a Semantic Pixel-Level Adaptation Transform approach to detector adaptation that efficiently generates cross-domain image pairs, and an adaptation method that exploits the continuity between gradually varying domains by adapting in sequence from the source to the most similar target domain. The models can be applied in a variety of visual recognition and prediction settings. They show new state-of-the-art results across multiple adaptation tasks.