This is the codes for the TNNLS paper Boosting Zero-shot Learning via Contrastive Optimization of Attribute Representations arxiv.
Dataset: please download the dataset, i.e., CUB, AWA2, SUN, and put it under datasets/
- apex
- PyTorch==1.7
- timm
or you can build the environment by docker pull dythu/zsl
see train_cub.sh