Abstract: Many approaches in generalized zero-shot learning rely on cross-modal mapping between the image feature space and the class embedding space. As labeled images are expensive, one direction is ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results