Describing Videos by Exploiting Temporal Structure

被引:670
作者
Yao, Li [1 ]
Torabi, Atousa [1 ]
Cho, Kyunghyun [1 ]
Ballas, Nicolas [1 ]
Pal, Christopher [2 ]
Larochelle, Hugo [3 ]
Courville, Aaron [1 ]
机构
[1] Univ Montreal, Montreal, PQ H3C 3J7, Canada
[2] Ecole Polytech, Montreal, PQ, Canada
[3] Univ Sherbrooke, Sherbrooke, PQ J1K 2R1, Canada
来源
2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV) | 2015年
关键词
D O I
10.1109/ICCV.2015.512
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application for video description. However, while images are static, working with videos requires modeling their dynamic temporal structure and then properly integrating that information into a natural language description. In this context, we propose an approach that successfully takes into account both the local and global temporal structure of videos to produce descriptions. First, our approach incorporates a spatial temporal 3-D convolutional neural network (3-D CNN) representation of the short temporal dynamics. The 3-D CNN representation is trained on video action recognition tasks, so as to produce a representation that is tuned to human motion and behavior. Second we propose a temporal attention mechanism that allows to go beyond local temporal modeling and learns to automatically select the most relevant temporal segments given the text-generating RNN. Our approach exceeds the current state-of-art for both BLEU and METEOR metrics on the Youtube2Text dataset. We also present results on a new, larger and more challenging dataset of paired video and natural language descriptions.
引用
收藏
页码:4507 / 4515
页数:9
相关论文
共 42 条
[1]  
[Anonymous], 2002, ACL
[2]  
[Anonymous], 2015, CVPR
[3]  
[Anonymous], 2010, P PYTHON SCI COMPUTI
[4]  
[Anonymous], 2013, PAMI
[5]  
[Anonymous], 2014, COLING
[6]  
[Anonymous], WORKSHOP DEEP LEARNI
[7]  
[Anonymous], ACL
[8]  
[Anonymous], 2015, ICML
[9]  
[Anonymous], 2015, ARXIV150204681
[10]  
[Anonymous], 2015, P 3 INT C LEARN REPR