Class knowledge overlay to visual feature learning for zero-shot image classification

Cheng Xie, Ting Zeng, Hongxin Xiang, Keqin Li, Yun Yang, Qing Liu

Abstract: Highlights•We propose a novel ZSL model that outperforms lots of models on several benchmarks.•We propose CKO method to gain more knowledge that improves performance of the model.•We re-train the ZSL model using SSL to reinforces category training and predicting.