VALSE 首页 活动通知 好文作者面授招 查看内容

20170809-16 徐迅:Semantic Spaces for Zero-Shot Behavior Analysis

2017-8-4 18:29| 发布者: 程一-计算所| 查看: 3811| 评论: 0

摘要: 报告嘉宾:徐迅(National University of Singapore)报告时间:2017年8月9日(星期三)晚20:00(北京时间)报告题目:Semantic Spaces for Zero-Shot Behavior Analysis主持人: 高常鑫(华中科技大学)报告摘要:T ...

报告嘉宾:徐迅(National University of Singapore)


报告题目:Semantic Spaces for Zero-Shot Behavior Analysis 

主持人:  高常鑫(华中科技大学) 


The number of categories for behavior analysis, e.g., action recognition and crowd analysis, is growing rapidly and it has become increasingly hard to label sufficient training data for learning conventional models for all categories. Instead of collecting ever more data and labelling them exhaustively for all categories, an attractive alternative approach is “zero-shot learning” (ZSL).  To this end, we construct a mapping between visual features and a semantic descriptor of each action category, allowing new categories to be recognized in the absence of any visual training data. Moreover, we note each dimension of this mapping is usually trained in an isolated way which ignores the correlation between dimensions. To tackle this issue, we explore a multi-task visual to semantic embedding and interestingly discover a lower dimension embedding in which nearest neighbor matching is more meaningful. Finally, we are aware that there is an inherent miss-match of semantic embedding learned from text and visual cues. To mitigate this gap, we exploit the statistic co-occurrence of multi-label attributes to refine the textual semantic embedding.


[1] Xun Xu, Timothy Hospedales and Shaogang Gong, Transductive Zero-Shot Action Recognition by Word-Vector Embedding, International Journal of Computer Vision (IJCV), 2017

[2] Xun Xu, Timothy Hospedales and Shaogang Gong, Multi-Task Zero-Shot Action Recognition with Prioritised Data Augmentation, European Conference on Computer Vision (ECCV), 2016

[3] Xun Xu, Shaogang Gong and Timothy Hospedales, Zero-Shot Crowd Behaviour Recognition, In Murino, Shah, Cristani, Savarese (Eds.), Group and Crowd Behaviour Understanding in Computer Vision , Elsevier, April 2017.


徐迅,现于新加坡国立大学开展博士后研究,于2016年获得英国伦敦大学玛丽女王学院计算机科学博士学位。目前主要从事监控场景理解,行为识别,迁移学习,零样本学习和运动分割的研究,在IJCV/ECCV/IEEE Transactions on CSVT等期刊和会议发表多篇论文。






小黑屋|手机版|Archiver|Vision And Learning SEminar

GMT+8, 2021-4-14 06:28 , Processed in 0.015584 second(s), 15 queries .

Powered by Discuz! X3.4

Copyright © 2001-2020, Tencent Cloud.