HOU Yuenan (侯跃南)

Yuenan Hou is a 4th-year Ph.D. candidate at the Department of Information Engineering, the Chinese University of Hong Kong. He is studying in the Multimedia Lab, under the supervision of Prof. Chen Change Loy and Prof. Xiaoou Tang.

He received his B.E degree from Nanjing University in July 2017, supervised by Prof. Chunlin Chen and Prof. Xianzhong Zhou. After that, he worked with Dr. Chunxiao Liu and Dr. Zheng Ma during his internship at Sensetime Research.

Currently, he is working in the project of GAN compression. His research interest mainly lies in model compression, e.g., knowledge distillation and network pruning.

Email / LinkedIn / Google Scholar / Github / CV

News
  • One paper is accepted by MICCAI 2021! (2021-06-12) NEW!

  • One paper is accepted by CVPR 2020! (2020-02-24)

  • One paper is accepted by ICCV 2019! (2019-07-23)

  • One paper is accepted by AAAI 2019 as Oral! (2018-11-23)

Education
cuhk

Aug. 2017 - Jul. 2021 (Expected), Department of Information Engineering, the Chinese University of Hong Kong

Doctor of Philosophy

nanjing

Sep. 2013 - Jul. 2017 , Department of Automation, Nanjing University

Bachelor of Engineering

Industrial Experience
sensetime

Apr. 2020 - May. 2020 , internship in SenseTime

Team leader: Dr. Zhe Wang

Research direction: knowledge distillation and network pruning in 3d detection

sensetime

Mar. 2017 - May. 2017 , internship in SenseTime

Team leader: Dr. Chunxiao Liu and Dr. Zheng Ma

Research direction: reinforcement learning in autonomous driving

Publications
CRCKD

Categorical Relation-Preserving Contrastive Knowledge Distillation for Medical Image Classification
Xiaohan Xing, Yuenan Hou, Hang Li, Yixuan Yuan, Hongsheng Li, Max Q.-H. Meng
International Conference on Medical Image Computing and Computer Assisted Intervension (MICCAI), 2021
[pdf] [code]

We propose a novel Categorical Relation-preserving Contrastive Knowledge Distillation (CRCKD) algorithm to distill the relational knowledge more comprehensively from the teacher model.

IntRA-KD

Inter-Region Affinity Distillation for Road Marking Segmentation
Yuenan Hou, Zheng Ma, Chunxiao Liu, Tak-Wai Hui, Chen Change Loy
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020
[pdf] [code] GitHub stars GitHub forks

We present a novel knowledge distillation approach, i.e., Inter-Region Affinity KD (IntRA-KD), which can transfer ‘knowledge’ on scene structure more effectively from a teacher to a student model.

SAD

Learning Lightweight Lane Detection CNNs by Self Attention Distillation
Yuenan Hou, Zheng Ma, Chunxiao Liu, Chen Change Loy
International Conference on Computer Vision (ICCV), 2019
[pdf] [code] GitHub stars GitHub forks

We present a novel knowledge distillation approach, i.e., Self Attention Distillation (SAD), which allows a model to learn from itself and gains substantial improvement without any additional supervision or labels.

FM-Net2018

Learning to Steer by Mimicking Features from Heterogeneous Auxiliary Networks
Yuenan Hou, Zheng Ma, Chunxiao Liu, Chen Change Loy
AAAI Conference on Artificial Intelligence (AAAI, oral), 2019
[pdf] [code] GitHub stars GitHub forks

We considerably improve the accuracy and robustness of our steering angle predictive model by distilling multi-layer knowledge from multiple heterogeneous auxiliary networks.

per_ddpg

A Novel DDPG Method with Prioritized Experience Replay
Yuenan Hou, Lifeng Liu, Qing Wei, Xudong Xu, Chunlin Chen
IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2017
[pdf] [code] GitHub stars GitHub forks

We proposed a prioritized experience replay method for the DDPG algorithm, where prioritized sampling is adopted instead of uniform sampling.

Preprint
PEEL

Network Pruning via Resource Reallocation
Yuenan Hou, Zheng Ma, Chunxiao Liu, Zhe Wang, Chen Change Loy
arXiv preprint arXiv:2103.01847, 2021
[pdf] [code]

We propose a simple yet effective channel pruning technique, termed network Pruning via rEsource rEalLocation (PEEL), to quickly produce a desired slim model via conducting resource reallocation on a predefined backbone.

ENet_label

Agnostic Lane Detection
Yuenan Hou
arXiv preprint arXiv:1905.03704, 2019
[pdf] [code] GitHub stars GitHub forks

We released a lightweight lane detection model, i.e., ENet-label, which can detect an arbitrary number of lanes and extremely thin lanes at 50 fps in theory.

Projects
  • GAN Compression, Dec. 2020 - Now

  • Network Pruning, Mar. 2020 - Nov. 2020

  • Road Marking Segmentation, Jun. 2019 - Feb. 2020

Selected course work
Academic Services
  • Conference reviewer of CVPR, ICCV, AAAI

  • Journal reviewer of TPAMI, IJCV, TIP, Neurocomputing, RA-L, IET, TGRS, JSTARS

  • Invited talk at School of Computer Science, Wuhan University, "Improving Deep Network Performance via Model Compression", 2021 [ppt pdf]

  • Invited talk at School of Cyber Science and Engineering, Wuhan University, "Improving Deep Network Performance via Model Compression", 2021

  • Invited talk at School of Artificial Intelligence, Sun Yat-sen University, "Improving Deep Network Performance via Model Compression", 2020

Honors and Awards
  • Rank 1st in ApolloScape Lane Segmentation Challenge, Baidu, 1/104, the "Codes-for-IntRA-KD" entity, Sep. 2020

  • Postgraduate Scholarship, the Chinese University of Hong Kong, 2017 ~ now

  • Third Prize of Jiangsu Province for Undergraduate Thesis, Nanjing University, 2017 [paper pdf] [awards list]

  • Zhenggang Scholarship, Nanjing University, 2017

  • Mathematical Contest in Modeling (MCM), Meritorious Winner (13%), 2016 [paper pdf]

  • Liao's Scholarship, Nanjing University, 2015

  • National Scholarship, Nanjing University, 2014