Generalized Zero-Shot Learning (GZSL) targets recognizing new categories bylearning transferable image representations . Existing methods find that, by aligning image representations with corresponding semantic labels, thesemantic-aligned representations can be transferred to unseen categories . In this paper, we propose a novel Dual-ContrastiveEmbedding Network (DCEN) that simultaneously learns task-specific and task-independent knowledge via semantic alignment and instance discrimination . DCEN leverages task labels to cluster representations of the samesemantic category by cross-modal contrastive learning and exploringsemantic-visual complementarity. DCEN obtainsaveraged a 4.1% improvement on four public benchmarks, according to the paper . Compared to high-level seen category supervision, this instancediscrimination

Author(s) : Chaoqun Wang, Xuejin Chen, Shaobo Min, Xiaoyan Sun, Houqiang Li

Links : PDF - Abstract

Code :

https://github.com/oktantod/RoboND-DeepLearning-Project


Coursera

Keywords : representations - task - learning - dcen - knowledge -

Leave a Reply

Your email address will not be published. Required fields are marked *