Abstract
7 Generalized zero-shot learning (GZSL) is a challenging class of vision and knowl8 edge transfer problems in which both seen and unseen classes appear during 9 testing. Existing GZSL approaches either suffer from semantic loss and discard10 discriminative information at the embedding stage, or cannot guarantee the visual-11 semantic interactions. To address these limitations, we propose a Dual Adversarial12 Semantics-Consistent Network (referred to as DASCN), which learns both primal13 and dual Generative Adversarial Networks (GANs) in a unified framework for14 GZSL. In DASCN, the primal GAN learns to synthesize inter-class discriminative15 and semantics-preserving visual features from both the semantic representations of16 seen/unseen classes and the ones reconstructed by the dual GAN. The dual GAN17 enforces the synthetic visual features to represent prior semantic knowledge well18 via semantics-consistent adversarial learning. To the best of our knowledge, this19 is the first work that employs a novel dual-GAN mechanism for GZSL. Extensive20 experiments show that our approach achieves significant improvements over the21 state-of-the-art approaches.