XIONG, Yuanjun's Homepage

Mr.Xiong Yuanjun - 熊元骏

Currently working in a startup.

Seattle, Washington, U.S.

yjxiong at ie.cuhk.edu.hk

Mr. Xiong Yuanjun is currently working in a startup on artificial intelligence. He worked as a principal scientist at Amazon Rekognition from 2017 to 2021. He was previously a postdoctoral fellow in department of information engineering, CUHK, working closely with Prof. Lin Dahua and Prof. Tang Xiaoou. Yuanjun received his Ph.D. from the department of information engineering at CUHK in 2016 and B.Eng from department of automation at Tsinghua University in 2012.

[Detailed CV] [Google Scholar] [Github: yjxiong]

Recent Activities

ELODI, a cross model compatibility method that utilizes ensembles to achieve forward compatibility is accepted to TPAMI. [preprint] NEW!
Joined a startup on artificial intelligence.
Organized ICCV 2021 tutorial on cross-model compatibility [Tutorial link].
A new paper on architecture-level compatible visual representation has been released. See the paper link.
Two papers accepted to CVPR 2021. The positive congruent training paper is accepted as oral.
Our new paper on backward compatibility and regression testing of deep models are released on Arxiv. Stay tuned for more incoming papers in this new research direction: Positive-Congruent Training: Towards Regression-Free Model Updates.
An open-source implementation of our paper Towards Backward Compatible Representation Learning has been released on Github.
Three papers accepted to ECCV 2020.
Arxiv preprint for the paper Motion Guided 3D Pose Estimation from Videos is released. We present a new way for supervising 3D pose estimation models surpassing other state-of-the-art methods by large margins.
Arxiv preprint for the paper Omni-sourced Webly-supervised Learning for Video Recognition is relased with the highest accuracy by far on the Kinetics400 dataset. [Press Coverage].
Our paper on backward compatible representation learning is accepted to CVPR 2020. It opens up a new area for research with wide industrial applications. The preprint has been released on Arxiv.
The interactive demo of our CSGN paper is online. Happy composing your own action sequence!
The journal version of SSN is accepted to IJCV.
Two papers accepted to ICCV 2019.
We have open sourced MMAction as a part of the OpenMMLAB intiative to support action understanding research, with implementation of all our previous works (including TrajactoryNet at NIPS'18) and latest state-of-the-art methods for action recognition, temporal detection, and spatial-temporal detection.
I am co-organizing the WIDER Face and Pedestrain Challenge held in conjunction with ICCV2019. This year we have four tracks: face detection (a teaser challenge with runtime cost is introduced), pedestrain detection, cast search by portrait, and person search by language.
One paper accepted to NeurIPS2018.
The extended version of the TSN paper is accepted to T-PAMI.
I am co-organizing the WIDER Face and Pedestrain Challenge held in conjunction with CVPR2018. This year we have three tracks: face detection, pedestrain detection, and person search.
Code and models of our CVPR2018 paper on unsupervised learning are released.
Three papers accepted to CVPR2018.
Code and models for our ST-GCN paper at AAAI-18 are released.
One paper accepted to AAAI-18.
Joined Amazon Rekognition. Thrilled to start a new journey.
We released the code and models of SSN.
We released the PyTorch implementation of TSN.
The SSN temporal action detection paper is accepted to ICCV2017, see you in Venice!
One paper accepted to CVPR2017.
Project page of our TSN paper is online now. Check out the [Website].
Successfully defended my thesis "Visual Understanding by Learning from Multiple Data Aspects".
Code and models of temporal segment networks (TSN), ECCV 2016, are released.
One paper accepted to ECCV 2016.
Our team CUHK & ETHZ & SIAT won the first place in ActivityNet Large Scale Activity Recognition Challenge 2016! This year the challenge is hosted together with CVPR'16. [Code] and [Online Demo] released!
Baseline results and models on WIDER dataset released [Dataset Page].
Technical report of state-of-the-art performance on action recognition, [Arxiv article]. With the report we released the Caffe extension with Multi-GPU training support. [Code].
One paper accepted to CVPR 2015.
Visiting Researcher at Google from Jan 2015 - Apr 2015.
One paper accepted to NIPS 2014.
One paper accepted to ICCV 2013.
Started doing research on Computer Vision.

Selected Publications

Yue Zhao, Yantao Shen, Yuanjun Xiong, Shuo Yang, Wei Xia, Zhuowen Tu, Bernt Schiele, Stefano Soatto, ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Training, IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), 2024. [Preprint]

Zeyi Sun, Ye Fang, Tong Wu, Pan Zhang, Yuhang Zang, Shu Kong, Yuanjun Xiong, Dahua Lin, Jiaqi Wang, Alpha-CLIP: A CLIP Model Focusing on Wherever You Want, Computer Vision and Pattern Recognition, (CVPR), 2024. [Preprint]

Jiarui Cai, Mingze Xu, Yuanjun Xiong, Wei Xia, Zhuowen Tu, Stefano Soatto, MeMOT: Multi-Object Tracking with Memory, Computer Vision and Pattern Recognition, (CVPR), 2022. [Paper]

Mingze Xu, Yuanjun Xiong, Hao Chen, Xinyu Li, Wei Xia, Zhuowen Tu, Stefano Soatto, Long short-term transformer for online action detection, Advances on Neural Information Processing (NeurIPS), 2021. [Paper]

Rahul Duggal, Hao Zhou, Shuo Yang, Yuanjun Xiong, Wei Xia, Zhuowen Tu, Stefano Soatto, Compatibility-Aware Heterogeneous Visual Search, Computer Vision and Pattern Recognition (CVPR), 2021. [Preprint]

Sijie Yan, Yuanjun Xiong, Kaustav Kundu, Shuo Yang, Siqi Deng, Meng Wang, Wei Xia, Stefano Soatto, Positive-Congruent Training: Towards Regression-Free Model Updates, Computer Vision and Pattern Recognition (CVPR), 2021. [Preprint]

Haodong Duan, Yue Zhao, Yuanjun Xiong, Wentao Liu, Dahua Lin, Omni-sourced Webly-supervised Learning for Video Recognition, European Conference on Computer Vision (ECCV), Edinburg, 2020. [Preprint]

Jingbo Wang, Sijie Yan, Yuanjun Xiong, Dahua Lin, Motion Guided 3D Pose Estimation from Videos, European Conference on Computer Vision, (ECCV), Edinburg, 2020. [Preprint]

Guha Balakrishnan, Yuanjun Xiong, Wei Xia, Pietro Perona, Towards causal Benchmarking of Bias in Face Analysis Algorithms, European Conference on Computer Vision (ECCV), Edinburg, 2020. [Preprint]

Yantao Shen, Yuanjun Xiong, Wei Xia, Stefano Soatto, Towards Backward Compatible Representation Learning, Computer Vision and Pattern Recognition (CVPR), Seattle, U.S., 2020. [PDF][Code][Press Coverage]

Sijie Yan, Zhizhong Li, Yuanjun Xiong, Dahua Lin, Convolutional Sequence Generation for Skeleton-Based Action Synthesis, IEEE International Conference on Computer Vision (ICCV), 2019, Seoul, South Korea, 2019. [PDF][Code][Demo Video][Web Demo][MMD Dataset]

Brais Martínez, Davide Modolo, Yuanjun Xiong, Joseph Tighe, Action Recognition with Spatial-Temporal Discriminative Filter Banks, IEEE International Conference on Computer Vision (ICCV), 2019, Seoul, South Korea, 2019. [PDF]

Yue Zhao, Yuanjun Xiong, Limin Wang, Zhirong Wu, Dahua Lin, and Xiaoou Tang, Temporal Action Detection with Structure Segment Networks, International Journal of Computer Vision (IJCV). [PDF][Code & Models][Project Page]

Limin Wang, Yuanjun Xiong, Zhe Wang, Yu Qiao, Dahua Lin, Xiaoou Tang, and Luc Van Gool, Temporal Segment Networks for Action Recognition in Videos, IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), 2018. [PDF]

Yue Zhao, Yuanjun Xiong, Dahua Lin, Trajectory Convolution for Action Recognition, Neural Information Processing Systems (NeurIPS), Montreal, Canada, 2018. [PDF][Poster][Code]

Yue Zhao, Yuanjun Xiong, Dahua Lin, Recognizing Actions by Disentangling Components of Dynamics, Computer Vision and Pattern Recognition (CVPR), Salt Lake City, U.S., 2018. [PDF]

Zhirong Wu, Yuanjun Xiong, Stella Yu, Dahua Lin, Unsupervised Feature Learning via Non-parametric Instance-level Discrimination, Computer Vision and Pattern Recognition (CVPR), Salt Lake City, U.S., 2018. [PDF] [Code & Models]

Kai Chen, Jiaqi Wang, Shuo Yang, Xingcheng Zhang, Yuanjun Xiong,Chen Change Loy, Dahua Lin, Optimizing Video Object Detection via a Scale-Time Lattice, Computer Vision and Pattern Recognition (CVPR), Salt Lake City, U.S., 2018. [PDF]

Sijie Yan, Yuanjun Xiong, and Dahua Lin, Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition, The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18), New Orlean, U.S., 2018. [PDF][Code & Models]

Yue Zhao, Yuanjun Xiong, Limin Wang, Zhirong Wu, Dahua Lin, and Xiaoou Tang, Temporal Action Detection with Structure Segment Networks, IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 2017. [PDF][Code & Models][Project Page]

Limin Wang, Yuanjun Xiong, Dahua Lin, and Luc Van Gool, UntrimmedNets for Weakly Supervised Action Recognition and Detection, Computer Vision and Pattern Recognition (CVPR), Hawaii, U.S., 2017. [PDF][Code & Models]

Limin Wang, Sheng Guo, Weilin Huang, Yuanjun Xiong, Yu Qiao, Knowledge Guided Disambiguation for Large-Scale Scene Classification With Multi-Resolution CNNs, IEEE Transactions on Image Processing (TIP), 2017. [PDF]

Limin Wang, Yuanjun Xiong, Zhe Wang, Yu Qiao, Dahua Lin, Xiaoou Tang, and Luc Van Gool, Temporal Segment Networks: Towards Good Practices for Deep Action Recognition, European Conference on Computer Vision (ECCV), Amsterdam, Netherlands, 2016. [PDF][Supplementary] [Code & Models][Project Page]

Yuanjun Xiong, Kai Zhu, Dahua Lin, and Xiaoou Tang, Recognize Complex Events from Static Images by Fusing Deep Channels, Computer Vision and Pattern Recognition (CVPR), Boston, U.S., 2015. [PDF][Supplementary] [Dataset]

Yuanjun Xiong, Wei Liu, Deli Zhao, and Xiaoou Tang. Zeta Hull Pursuits: Learning Nonconvex Data Hulls, Advances in Neural Information Processing System (NIPS 27), Montreal, Canada, 2014. [PDF][Supplementary]

Yuanjun Xiong, Wei Liu, Deli Zhao, and Xiaoou Tang. Face Recognition via Archetype Hull Ranking, IEEE International Conference on Computer Vision (ICCV), Sydney, Australia, 2013.[PDF]