Abstract
Feature generating networks face to the most important question, which is the fitting difference (inconsistence) of the distribution between the generated feature and the real data. This inconsistence further influence the performance of the networks model, because training samples from seen classes is disjointed with testing samples from unseen classes in zero-shot learning (ZSL). In generalization zero-shot learning (GZSL), testing samples come from not only seen classes but also unseen classes for closer to the practical situation. Therefore, most of feature generating networks difficultly obtain satisfactory performance for the challenging GZSL by adversarial learning the distribution of semantic classes. To alleviate the negative influence of this inconsistence for ZSL and GZSL, transfer feature generating networks with semantic classes structure (TFGNSCS) is proposed to construct networks model for improving the performance of ZSL and GZSL. TFGNSCS can not only consider the semantic structure relationship between seen and unseen classes, but also learn the difference of generating features by transferring classification model information from seen to unseen classes in networks. The proposed method can integrate the transfer loss, the classification loss and the Wasserstein distance loss to generate enough CNN features, on which softmax classifiers are trained for ZSL and GZSL. Experiments demonstrate that the performance of TFGNSCS outperforms that of the state of the arts on four challenging datasets, which are CUB,FLO,SUN, AWA in GZSL.
Highlights
Based on large quantities of labeled training data, deep learning can capture the various patterns of data for large-scale recognition problems
Our contributions have three points as follows. (a) We present a novel adversarial generative model TFGNSCS that synthesizes the more discriminative CNN features of classes for zero-shot learning (ZSL) or generalized zero-shot learning (GZSL) by structure propagation enhancing model learning, which can integrates the transfer loss, the classification loss and the Wasserstein distance loss to balance the inconsistency between the generated feature and the real data. (b) In four challenging datasets with different size or granularity, the proposed TFGNSCS outperforms the state-of-theart models in the GZSL setting, and shows the important role of structure propagation to build adversarial generative model in ZSL or GZSL. (c) Our model is generalized by the consideration of different transfer information methods to evaluate their influence for the more discriminative features
We have proposed a transfer feature generating networks with semantic class structure (TFGNSCS) method to address the inconsistency between the generated feature and the real feature for improving the performance of ZSL and GZSL
Summary
Based on large quantities of labeled training data, deep learning can capture the various patterns of data for large-scale recognition problems. The model trained based on the classification loss of seen classes can not effectively adapt the distribution of unseen classes, and this situation causes the fitting difference (inconsistency) of the distribution between the generated feature and the real data. VOLUME 7, 2019 seen and unseen classes by structure propagation (the details in section II-B), and the other (the detail is defined by equation (4) in section III-A) balances the difference of generating features between seen and unseen classes by discriminator information. (a) We present a novel adversarial generative model TFGNSCS that synthesizes the more discriminative CNN features of classes for ZSL or GZSL by structure propagation enhancing model learning, which can integrates the transfer loss, the classification loss and the Wasserstein distance loss to balance the inconsistency between the generated feature and the real data. Our contributions have three points as follows. (a) We present a novel adversarial generative model TFGNSCS that synthesizes the more discriminative CNN features of classes for ZSL or GZSL by structure propagation enhancing model learning, which can integrates the transfer loss, the classification loss and the Wasserstein distance loss to balance the inconsistency between the generated feature and the real data. (b) In four challenging datasets with different size or granularity, the proposed TFGNSCS outperforms the state-of-theart models in the GZSL setting, and shows the important role of structure propagation to build adversarial generative model in ZSL or GZSL. (c) Our model is generalized by the consideration of different transfer information methods (structure propagation methods) to evaluate their influence for the more discriminative features
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
