TY - GEN
T1 - Co-consistent regularization with discriminative feature for zero-shot learning
AU - Tian, Yanling
AU - Zhang, Weitong
AU - Zhang, Qieshi
AU - Cheng, Jun
AU - Hao, Pengyi
AU - Lu, Gang
N1 - Funding Information:
Acknowledgments. This work was supported by National Natural Science Foundation of China (61772508, 61801428, U1713213), National Key R&D Program of China (2017YFB1402100), Zhejiang Provincial Natural Science Foundation (LY18F020034), Natural Science Basic Research Plan in Shaanxi Province of China (2017JM6101, 2017JM6060, 2017JQ6077, 2017JM6103), Guangdong Technology Project (2016B010108010, 2016B010125003, 2017B010110007), CAS Key Technology Talent Program, Shenzhen Engineering Laboratory for 3D Content Generating Technologies ([2017]476), Shenzhen Technology Project (JCYJ 20170413152535587, JSGG20160331185256983, JSGG20160229115709109), Key Laboratory of Human-Machine Intelligence-Synergy Systems, Shenzhen Institutes of Advanced Technology, CAS (2014DP173025), Fundamental Research Funds for the Central Universities (GK201703060, GK201801004), Teaching Reform and Research Project of Shaanxi Normal University (17JG33).
PY - 2018
Y1 - 2018
N2 - With the development of deep learning, zero-shot learning (ZSL) issues deserve more attention. Due to the problems of projection domain shift and discriminative feature extraction, we propose an end-to-end framework, which is different from traditional ZSL methods in the following two aspects: (1) we use a cascaded network to automatically locate discriminative regions, which can better extract latent features and contribute to the representation of key semantic attributes. (2) our framework achieves mapping in visual-semantic embedding space and calculation procedure of the dot product in deep learning framework. In addition, a joint loss function is designed for the regularization constraint of the whole method and achieves supervised learning, which enhances generalization ability in test set. In this paper, we make some experiments on Animals with Attributes 2 (AwA2), Caltech-UCSD Birds 200-2011 (CUB) and SUN datasets, which achieves better results compared to the state-of-the-art methods.
AB - With the development of deep learning, zero-shot learning (ZSL) issues deserve more attention. Due to the problems of projection domain shift and discriminative feature extraction, we propose an end-to-end framework, which is different from traditional ZSL methods in the following two aspects: (1) we use a cascaded network to automatically locate discriminative regions, which can better extract latent features and contribute to the representation of key semantic attributes. (2) our framework achieves mapping in visual-semantic embedding space and calculation procedure of the dot product in deep learning framework. In addition, a joint loss function is designed for the regularization constraint of the whole method and achieves supervised learning, which enhances generalization ability in test set. In this paper, we make some experiments on Animals with Attributes 2 (AwA2), Caltech-UCSD Birds 200-2011 (CUB) and SUN datasets, which achieves better results compared to the state-of-the-art methods.
KW - Discriminative region
KW - Projection domain shift
KW - Regularization
KW - Supervised learning
KW - Zero-shot learning (ZSL)
UR - http://www.scopus.com/inward/record.url?scp=85059090663&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85059090663&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-04167-0_4
DO - 10.1007/978-3-030-04167-0_4
M3 - Conference contribution
AN - SCOPUS:85059090663
SN - 9783030041663
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 33
EP - 45
BT - Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings
A2 - Cheng, Long
A2 - Leung, Andrew Chi Sing
A2 - Ozawa, Seiichi
PB - Springer Verlag
T2 - 25th International Conference on Neural Information Processing, ICONIP 2018
Y2 - 13 December 2018 through 16 December 2018
ER -