Yanzhao Zhou

Yanzhao Zhou (周彦钊)

Ph.D. Student in Computer Vision

About Me

I am a Ph.D. Student in Pattern Recognition and Intelligent System Development Laboratory (PriSDL) at University of Chinese Academy of Sciences (UCAS), advised by Prof. Jianbin Jiao. I got the B.Eng. in Communication Engineering at Beijing JiaoTong University in July 2015.

My research interests focus on Visual Perception (object recognition, localization, segmentation, etc.) and Multimodal Reasoning (vision + language). Particularly I am committed to developing deep learning systems that can learn from limited annotations like humans.



Learning Saliency Propagation for Semi-Supervised Instance Segmentation

Yanzhao Zhou, Xing Wang, Jianbin Jiao, Trevor Darrell, and Fisher Yu

IEEE Int'l Conf. on Computer Vision and Pattern Recognition (CVPR), 2020.

ShapeProp can benefit from more bounding box supervision to locate the instances more accurately and utilize the feature activations from the larger number of instances to achieve more accurate segmentation.

Paper Code & Demo Bibtex

Selective Sparse Sampling for Fine-grained Image Recognition

Yao Ding*, Yanzhao Zhou*, Yi Zhu, Qixiang Ye, and Jianbin Jiao

IEEE Int'l Conf. on Computer Vision (ICCV), 2019. *-indicates equal contributions

S3N learns a set of sparse attention from class peak responses for fine-grained object recognition.

Paper Code & Demo Bibtex

Deep Reason: A Strong Baseline for Real-World Visual Reasoning

Chenfei Wu, Yanzhao Zhou, Gen Li, Nan Duan, Duyu Tang, and Xiaojie Wang

IEEE Int'l Conf. on Computer Vision and Pattern Recognition (CVPR) GQA Workshop, 2019.

DREAM learns a strong multi-modal representation for visual question answering. DREAM achieves the top result on GQA benchmark.

Paper Bibtex

Learning Instance Activation Maps for Weakly Supervised Instance Segmentation

Yi Zhu, Yanzhao Zhou, Huijuan Xu, Qixiang Ye, David Doermann, and Jianbin Jiao

IEEE Int'l Conf. on Computer Vision and Pattern Recognition (CVPR), 2019.

IAM learns from image-level labels and noisy proposals to produce high-quality instance-aware visual cues.

Paper Bibtex

Weakly Supervised Instance Segmentation using Class Peak Response

Yanzhao Zhou, Yi Zhu, Qixiang Ye, Qiang Qiu, and Jianbin Jiao

IEEE Int'l Conf. on Computer Vision and Pattern Recognition (CVPR) spotlight, 2018.

The proposed technique PRM extracts fine-detailed instance-aware visual cues from DCNNs trained with image-level labels.

Paper Supp Poster Code & Demo Presentation Bibtex

Soft Proposal Networks for Weakly Supervised Object Localization

Yi Zhu, Yanzhao Zhou, Qixiang Ye, Qiang Qiu, and Jianbin Jiao

IEEE Int'l Conf. on Computer Vision (ICCV), 2017.

SPN is an end-to-end weakly supervised localization framework with nearly cost-free object proposal.

Paper Supp Code & Demo Bibtex

Oriented Response Networks

Oriented Response Networks

Yanzhao Zhou, Qixiang Ye, Qiang Qiu, and Jianbin Jiao

IEEE Int'l Conf. on Computer Vision and Pattern Recognition (CVPR), 2017.

ORN boosts the generalization ability of DCNNs to handle rotation problems.

Paper Supp Poster Code & Demo Bibtex

Selected Projects


Project Nest, 2018 Deep Learning Module Manager Python

In this project, I developed a flexible deep learning module manager, which aims at encouraging code reuse and sharing. It ships with a bunch of useful features, such as CLI based module management, runtime checking, and experimental task runner, etc. You can integrate Nest with any deep learning framework you like that provides a python interface.

Project page
Aerial Detection System

Aerial Detection System, 2017 Vision Aerial Image Detection

To address the challenge of object rotations in the aerial images, I led a team of four people to develop an Aerial Object Detection system based on my CVPR 2017 Paper (ORN). The system can be trained end-to-end with limited data and generate precise oriented bboxes. It won first place ( vehicle detection) and second place (plane detection) in the XingTu Cup Remote-Sensing Contest.

Media Coverage [1], [2]
CoalYard 3D Reconstruction System

CoalYard 3D Reconstruction System, 2016 Vision Radar 3D Modeling

In this project, I designed a software system that can reconstruct the 3D surface of coal heaps in real-time based on the data of microwave radar and vision sensors. The system was successfully applied to Huanghua Port, Hebei province, China.

Drone Delivery Guiding System

Drone Delivery Guiding System, 2015 Drone Tracking Binocular Vision

As the project leader, I developed a vision tracking system (hardware & software), which can detect approaching drones (multi-rotor UAVs) under various light conditions and locate their accurate 3D position (mm-level) in real-time (~60Hz). The system uses the tracking feedback to auto-pilot a drone to precisely deliver cargo.

Check Media Coverage Patent
Intelligent Robot Training System

Intelligent Robot Training System, 2014 Robot Vision Neural Networks

I led a team of three people to develop a Robot Auto-Training system that uses computer vision algorithms to analyze the trajectory errors of the robot and then automatically adjust the control loop via a BP-Net based method. In the demo, a robot mouse is trained to avoid obstacles in a maze.

Intelligent Traffic Light

Intelligent Traffic Light, 2013 Traffic Analysis Vision Optimization

To address the problem of urban traffic jam, I led a team of five people to develop a system that uses Optical Flow based vision algorithms to monitor the real-time intersection traffic and predict the trajectory of each vehicle. The system adjusts the traffic signal to maximize the passing rate according to the prediction.

Research Experience

Visiting Scholar - Berkeley Artificial Intelligence Research (BAIR) (07/2019 - 12/2019)

Mentor: Prof. Fisher Yu and Prof. Trevor Darrell

  • Research of semi-supervised instance segmentation.

Research Intern - Microsoft Research Asia (12/2018 - 07/2019)

Mentor: Dr. Nan Duan and Dr. Duyu Tang

  • Research of multimodal reasoning (vision + natural language).

Research Student - University of Chinese Academy of Sciences, PriSDL (09/2015 - Present)

Advised by Prof. Jianbin Jiao and co-advised by Prof. Qixiang Ye

  • Research of improving the capability of Convolutional Neural Networks (CNNs) to handle significant image rotations.
  • Research of developing weakly supervised object localization frameworks to process large-scale rough annotated data.

Visiting Student - Tsinghua University (12/2014 - 02/2015)

Advisor: Prof. Guiming Luo

  • Research of Multifunction Vehicle Bus Protocol software simulation.

Student - Beijing JiaoTong University (09/2011 - 07/2015)

  • Vision Anti-Thief System based on face recognition, 2012. (project leader)
  • Intelligent Traffic Light based on genetic algorithm, 2013. (project leader)
  • Intelligent Robot Training System based on neural networks, 2014. (project leader)
  • Drone Delivery Guiding System based on dual-camera visual tracking, 2015. (project leader)