eduzhai > Applied Sciences > Engineering >

Learning the Redundancy-free Features for Generalized Zero-Shot Object Recognition

  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Zero-shot object recognition or zero-shot learning aims to transfer theobject recognition ability among the semantically related categories, such asfine-grained animal or bird species. However, the images of differentfine-grained objects tend to merely exhibit subtle differences in appearance,which will severely deteriorate zero-shot object recognition. To reduce thesuperfluous information in the fine-grained objects, in this paper, we proposeto learn the redundancy-free features for generalized zero-shot learning. Weachieve our motivation by projecting the original visual features into a new(redundancy-free) feature space and then restricting the statistical dependencebetween these two feature spaces. Furthermore, we require the projectedfeatures to keep and even strengthen the category relationship in theredundancy-free feature space. In this way, we can remove the redundantinformation from the visual features without losing the discriminativeinformation. We extensively evaluate the performance on four benchmarkdatasets. The results show that our redundancy-free feature based generalizedzero-shot learning (RFF-GZSL) approach can outperform the state-of-the-artsoften by a large margin. Our code is available.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×