Zero-Shot Learning via Discriminative Dual Semantic Auto-Encoder

Zero-shot learning (ZSL) is an effective method to perform the recognition task without any training samples of specific classes. Most existing ZSL models put emphasis on embedding between visual space and semantic directly. However, few research whether human-designed features are discriminative enough recognize different Moreover, one-way mapping suffers from project domain shift problem. In this article, we propose learn a Discriminative Dual Semantic Auto-encoder (DDSA) based encoder-decoder paradigm solve DDSA attempts construct two bidirectional embeddings connect with help learned aligned which includes information features. Based DDSA, additionally Deep capture deep that more conducive zero-shot classification. The key proposed framework it implicitly exact principal features, not only semantic-preserving but also discriminative. Extensive experiments five benchmarks (SUN, CUB, AWA1, AWA2 aPY) demonstrate effectiveness state-of-the-art performance obtained both conventional generalized settings.