Auto‐encode the synthesis pseudo features for generalized zero‐shot learning

Zero-shot learning (ZSL) is to identify target categories without labeled data, in which semantic information used transfer knowledge from some seen categories. In the existing Generalized Zero-Shot Learning (GZSL) methods, domains shift problem always appeared during generating feature stage. order solve this problem, a new method Auto-Encode Synthesis Pseudo Features for GZSL task (AESPF-GZSL) proposed manuscript. Specifically, AESPF-GZSL trains generated features under auto-encoder framework and exploits attention mechanism train again. Then, are input classifier. The performed on three benchmark data sets referred as AWA, CUB SUN. experimental results show that achieves state-of-the art classifier accuracy both ZSL settings. setting, classification of our superior compared algorithms, improved by 0.40% AWA 0.30% SUN, respectively. And comparison algorithm 0.41% Harmonic mean 1.01%, 0.62%, 1.05% training set, testing harmonic average