Learning semantic double-autoencoder with attribute constraint for zero-shot recognition

Abstract

Existing zero-shot recognition (ZSR) approaches generally learn a projection function from the labelled training (source) dataset. However, applying the learned projection function without adaptation to the test (target) dataset is prone to the domain shift problem. In this paper, we propose a semantic double-autoencoder with attribute constraint (SDAWAC) mechanism to overcome the problem effectively. Specifically, we take the semantic encoder-decoder paradigm to learn a projection function in the source and target domains simultaneously. In addition, we introduce one constraint on source domain attributes into this work to improve the performance of our model. The experimental results on three benchmark datasets demonstrate the efficacy of our proposed method.