Jointly Modeling Deep Video and Compositional Text to Bridge Vision and Language in a Unified Framework

Abstract

Recently, joint video-language modeling has been attracting more and more attention. However, most existing approaches focus on exploring the language model upon on a fixed visual model. In this paper, we propose a unified framework that jointly models video and the corresponding text sentences. The framework consists of three parts: a compositional semantics language model, a deep video model and a joint embedding model. In our language model, we propose a dependency-tree structure model that embeds sentence into a continuous vector space, which preserves visually grounded meanings and word order. In the visual model, we leverage deep neural networks to capture essential semantic information from videos. In the joint embedding model, we minimize the distance of the outputs of the deep video model and compositional language model in the joint space, and update these two models jointly. Based on these three parts, our system is able to accomplish three tasks: 1) natural language generation, and 2) video retrieval and 3) language retrieval. In the experiments, the results show our approach outperforms SVM, CRF and CCA baselines in predicting Subject-Verb-Object triplet and natural sentence generation, and is better than CCA in video retrieval and language retrieval tasks.

Video Captioning with Listwise Supervision

“…There are generally two categories of methods for video captioning: template-based models (Kojima, Tamura, and Fukunaga 2002;Rohrbach et al 2013;Guadarrama et al 2013;Xu et al 2015) and sequence learning models (e.g., RNN) (Donahue et al 2015;Pan et al 2016a;Venugopalan et al 2015a;Yao et al 2015;Venugopalan et al 2015b;Pan et al 2016b). The former predefines the special rule for language grammar and then parses the sentence into several parts (e.g., subject, verb, object).…”

“…In (Guadarrama et al 2013), Guadarrama et al utilize semantic hierarchies to choose an appropriate level of the specificity and accuracy of sentence fragments. Recently, Xu et al design a unified framework in (Xu et al 2015), which consists of a compositional semantics language model, a deep video model and an embedding model to capture the joint video-language relationship for video sentence generation.…”

Egocentric Video Search via Physical Interactions

“…Many multi-modal retrieval frameworks that retrieve images using text queries (Jeon, Lavrenko, and Manmatha 2003;Guillaumin et al 2009;Xu et al 2015) have been proposed. However, we use the PCCA-based retrieval model, which easily combines motion and video features and can be easily extended to the re-ranking model.…”

Learning to Answer Questions from Image Using Convolutional Neural Network

“…Recently, the multimodal learning between image and language (Ma et al 2015;Nakamura et al 2013;Xu et al 2015b) has become an increasingly popular research area of artificial intelligence (AI). In particular, there have been rapid progresses on the tasks of bidirectional image and sentence retrieval (Frome et al 2013;Socher et al 2014;Klein et al 2015;Karpathy, Joulin, and Li 2014;Ma et al 2015;Ordonez, Kulkarni, and Berg 2011), and automatic image captioning (Chen and Zitnick 2014;Donahue et al 2014;Fang et al 2014;Kiros, Salakhutdinov, and Zemel 2014a;Klein et al 2015;Mao et al 2014a;Vinyals et al 2014;Xu et al 2015a).…”