CyCADA: Cycle-Consistent Adversarial Domain Adaptation

TL;DR

CyCADA introduces a cycle-consistent adversarial approach for domain adaptation that improves pixel-level and feature-level transfer without needing aligned image pairs, achieving state-of-the-art results.

Contribution

The paper presents a novel cycle-consistent adversarial model that enhances domain adaptation by combining pixel and feature space alignment without requiring aligned data.

Findings

Achieved state-of-the-art results in digit classification.

Improved semantic segmentation in synthetic to real domain transfer.

Demonstrated effectiveness across multiple visual recognition tasks.

Abstract

Domain adaptation is critical for success in new, unseen environments. Adversarial adaptation models applied in feature spaces discover domain invariant representations, but are difficult to visualize and sometimes fail to capture pixel-level and low-level domain shifts. Recent work has shown that generative adversarial networks combined with cycle-consistency constraints are surprisingly effective at mapping images between domains, even without the use of aligned image pairs. We propose a novel discriminatively-trained Cycle-Consistent Adversarial Domain Adaptation model. CyCADA adapts representations at both the pixel-level and feature-level, enforces cycle-consistency while leveraging a task loss, and does not require aligned pairs. Our model can be applied in a variety of visual recognition and prediction settings. We show new state-of-the-art results across multiple adaptation tasks, including digit classification and semantic segmentation of road scenes demonstrating transfer from synthetic to real world domains.