研究方向

(一)受脑启发的深度神经网络技术

 

连续学习Continual Learning

  • 统一单一任务连续学习 

成果1:小样本类增量连续学习 (CVPR-20 Oral)  [PDF] [Codes] [Slides][知乎]

 Xiaoyu Tao, Xiaopeng Hong, Xinyuan Chang, Songlin Dong, Xing Wei, Yihong Gong. Few-Shot Class-Incremental Learning. Proceedings of the 2020 Conference on Computer Vision and Pattern Recognition, CVPR20, 2020.

 

成果2:“温故而知新”双目标优化的实例增量连续学习(AAAI-20)  [PDF] [Codes] [Slides]

Xiaoyu Tao, Xiaopeng Hong, Xinyuan Chang, Yihong Gong. Bi-objective Continual Learning: Learning ‘New’ while Consolidating ‘Known,’ Proceedings of the AAAI Conference on Artificial Intelligence, AAAI20, 2020.

 

 成果3:拓扑保持的类增量连续学习 (ECCV-20)  [PDF] [Codes] [Slides][知乎]

 Xiaoyu Tao, Xiaopeng Hong, Xinyuan Chang, Songlin Dong, Xing Wei, Yihong Gong. Topology-Preserving Class-Incremental Learning. ECCV20, 2020.

 

类脑神经网络设计

 

  • 双通道神经网络 

成果1:类比-细节双通道神经网络 (CVPR-20 Oral)  [PDF] [Codes] [Slides][知乎]

 

 X. Tao, X. Hong, W. Shi, X. Chang, Y. Gong. Analogy-Detail Networks for Object Recognition. IEEE TNNLS, 2020.

 

(二)视频监控 (Visual Surveillence)

  • 多摄像头多目标跟踪

成果1:基于受限非负矩阵分解的跨摄像头多目标跟踪 IEEE TIP-20, ECCV Visdrone20 Chanllenge 冠军IEEE CVPR20 AI-CITY Challenge Runner-up) [PDF] [Codes] [Slides][知乎]

Y. He, X. Wei, X. Hong, W. Shi and Y. Gong, "Multi-Target Multi-Camera Tracking by Tracklet-to-Target Assignment," IEEE Transactions on Image Processing, vol. 29, pp. 5191-5205, 2020. DOI: 10.1109/TIP.2020.2980070

 

  • 目标计数、人群计数

成果1:基于贝叶斯损失(Bayesian Loss)的人群计数算法 (ICCV-19 Oral)  [PDF] [Codes] [Slides] [知乎]

 

 

Zhiheng Ma, Xing Wei, Xiaopeng Hong, Yihong Gong. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 6142-6151

Bibtex

@InProceedings{Ma_2019_ICCV,

author = {Ma, Zhiheng and Wei, Xing and Hong, Xiaopeng and Gong, Yihong},

title = {Bayesian Loss for Crowd Count Estimation With Point Supervision},

booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},

year = {2019}

}

 

成果2:从点中学到尺度:一种尺度可知的人群计数概率模型  (ACM MM20)[PDF] [Codes] [Slides][知乎]

Z. Ma, X. Wei, X. Hong, Y. Gong. Learning Scales from Points: A Scale-aware Probabilistic Model for Crowd Counting.  ACM Multimedia, 2020.

 

  • (跨模态)行人重识别

成果1:基于协同注意力机制提升的红外-可见光跨模态行人重识别(ACM MM20) [PDF] [Codes] [Slides][知乎]

 

X. Wei, D. Li, X. Hong, W. Ke, Y. Gong. Co-Attentive Lifting for Infrared-Visible Person Re-Identification. ACM Multimedia, 2020.

 

Bibtex

@inproceedings{wei_coattentive_reid20,

author={Wei, Xing and Li, Diangang and Hong, Xiaopeng and Ke, Wei and Gong, Yihong},

title={Co-Attentive Lifting for Infrared-Visible Person Re-Identification},

booktitle={Proceedings of the ACM Multimedia},

year={2020}

}

成果2:基于X模态的红外-可见光跨模态行人重识别 (AAAI-20) [PDF] [Codes] [Slides][知乎]

 

Diangang Li, Xing Wei, Xiaopeng Hong, Yihong Gong. Infrared-Visible Cross-Modal Person Re-Identification with an X Modality. Proceedings of the AAAI Conference on Artificial Intelligence, AAAI20, 2020.

Bibtex

@inproceedings{li_xiv_reid20,

author={Li, Diangang and Wei, Xing and Hong, Xiaopeng and Gong, Yihong},

title={Infrared-Visible Cross-Modal Person Re-Identification with an X Modality},

booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},

month = {February},

year={2020}

}

 

成果3:基于直推式学习的半监督行人重识别(PR 20) [PDF] [Codes] [Slides][知乎]

X. Chang, Z. Ma, X. Wei, X. Hong, Y. Gong. Transductive Semi-Supervised Metric Learning for Person Re-identification. Pattern Recogntion, 2020.

  • 姿态识别与行为分析

成果1:基于网络结构搜索的姿态识别 (AAAI-20) [PDF] [Codes] [Slides][知乎]

 

(三)策略分配与任务调度 (Scheduling and Allocation)

 

(四)缺陷检测(DEFECT INSPECTION

 

(五)深度学习模型结构与脆弱性分析(MODEL STRUCTURE & VULNERABILITY AND ROBUSTNESS

 

 

 

 

 

(六)脸部微小动作分析(FACIAL SUBTLE ACTION ANALYSIS)

 

 

 

成果1:自发微表请检测与识别方法的对比研究(IEEE TAFFC2018) [PDF] [Codes] [Slides][Reports:MIT Technology Review][Reports:DailyMail]

Xiaobai Li, Xiaopeng Hong, Antti Moilanen, Xiaohua Huang, Thomas Pfister, Guoying Zhao, Matti Pietikäinen. Towards Reading Hidden Emotions: A Comparative Study of Spontaneous Micro-expression Spotting and Recognition Methods. IEEE Transactions on Affective Computing, Vol. 9, No. 4, pp. 563-577, IEEE TAFFC, 2018.

 

在本文中,我们提出了一个基于运动放大和时间插值预处理方法,用快速HIGO-TOP描述子提取视频特征的微表情识别框架。该方法使用尽管是传统手工特征的框架,然而其在SMIC和CASME II数据库上的性能在15年到18年一直保持领先,直到19年初才被基于递归卷积神经网络方法超越(见成果3)。该文得到包括麻省理工学院技术综述MIT Technology Review和每日邮报DailyMail的专文报道。运动放大的代码可以在William T. Freeman教授课题组下载,时间插值模块的MATLAB代码可以点这里。快速LBP-TOP模块的代码点这里

 

成果2:基于黎曼流形的瞬时脸部运动表征方式 (IEEE TOMM2019) [PDF] [Codes] [Slides]

 X. Hong; W. Peng; M. Harandi; Z. Zhou; M. Pietikäinen, and G. Zhao. Characterizing Subtle Facial Movements via Riemannian Manifold.

 ACM Transactions on Multimedia Computing Communications and Applications, Vol. 15, No. 3s, pp. 1-24, ACM TOMM, 2019.

 

在本文中,我们做了两方面的工作。首先,将成果1中的运动放大和时间插值模块整合成一个统一的线性框架,不仅加快了预处理的速度,而且减少了中间视频存储带来的精度损失从而提高了性能。其次,我们提出了基于黎曼流形差分描述子的瞬时脸部运动表征方式。

 

成果3:基于递归卷积神经网络的自发微表情识别 (IEEE TMM2019) [PDF] [Codes] [Slides]

 Z. Xia; X. Hong; X. Gao; X. Feng; G. Zhao.  Spatiotemporal Recurrent Convolutional Networks for Recognizing Spontaneous Micro-expressions.
IEEE Transactions on Multimedia, Vol. 22, No. 3, pp. 626-640, IEEE TMM, 2019.

 

成果4:技术挑战赛和研讨会 (IEEE FG2018-Present) [挑战赛总结2020] [挑战赛总结2019] [挑战赛总结2018]