Learning visual and textual representations for multimodal matching and classification

PATTERN RECOGNITION

卷:84

Article

Liu, Yu¹  Liu, Li^2,3  Guo, Yanming²  Lew, Michael S.¹ 

[1] Leiden Univ, Dept Comp Sci, NL-2333 CA Leiden, Netherlands

[2] Natl Univ Def Technol, Coll Syst Engn, Changsha 410073, Hunan, Peoples R China

[3] Univ Oulu, Ctr Machine Vis & Signal Anal, Oulu 8000, Finland

关键词: Vision and language;    Multimodal matching;    Multimodal classification;    Deep learning;   

DOI  :  10.1016/j.patcog.2018.07.001

来源: Elsevier

【 摘 要 】

Multimodal learning has been an important and challenging problem for decades, which aims to bridge the modality gap between heterogeneous representations, such as vision and language. Unlike many current approaches which only focus on either multimodal matching or classification, we propose a unified network to jointly learn multimodal matching and classification (MMC-Net) between images and texts. The proposed MMC-Net model can seamlessly integrate the matching and classification components. It first learns visual and textual embedding features in the matching component, and then generates discriminative multimodal representations in the classification component. Combining the two components in a unified model can help in improving their performance. Moreover, we present a multi-stage training algorithm by minimizing both of the matching and classification loss functions. Experimental results on four well-known multimodal benchmarks demonstrate the effectiveness and efficiency of the proposed approach, which achieves competitive performance for multimodal matching and classification compared to state-of-the-art approaches. (C) 2018 Published by Elsevier Ltd.

【 授权许可】

Free