Fan, Lei

Fan Lei
Graduate student
School of Data and Computer Science, Sun Yat-sen University
No. 132, Waihuan East Road
Guangzhou, P.R. China


Sun Yat-sen University
9/2017 - 6/2019

Institute of Unmanned Systems, School of Data and Computer Science
Master of Engineering, expected in 6/2019
Supervised under Prof. Long Chen (Homepage)
Overall Ranking: 1/390
Sun Yat-sen University
9/2013 - 6/2017

School of Data and Computer Science
Major in Software Engineering
Bachelor of Engineering, received in 6/2017
Overall GPA: 3.8/4.0

Research Interests

Stereo vision
3D reconstruction
Semantic segmentation
Autonomous driving


L. Chen, L. Sun, T. Yang, L. Fan, K. Huang and Z. X,"RGB-T SLAM: A Flexible SLAM Framework by Combining Appearance and Thermal Information", accepted by IEEE International Conference on Robotics and Automation 2017. [pdf]
L. Chen, L. Fan, G. Xie, K. Huang and A. Nuchter, "Moving-Object Detection from Consecutive Stereo Pairs using Slanted Plane Smoothing", accepted by IEEE Transactions on Intelligent Transportation Systems. [pdf]
L. Chen, Y. He, and L. Fan "Let the Robot Tell: Describe Car Image with Natural Language via LSTM", accepted by Pattern Recognition Letters. [pdf]
L. Chen, L. Fan, J. Chen, D. Cao, and F. Wang, "A Full Density Stereo Matching System Based on the Combination of CNNs and Slanted-planes", accepted by IEEE Transactions on Systems, Man, and Cybernetics: Systems. [pdf]
L. Fan, L. Chen, K. Huang and D. Cao. "Planecell: Representing Structural Space with Plane Elements", accepted as Best Student Paper by IEEE Intelligent Vehicles Symposium 2018. [pdf]
L. Fan, L. Chen, C. Zhang, W. Tian and D. Cao "Collaborative 3D Completion of Color and Depth in a Specified Area with Superpixels" by IEEE Transactions on Industrial Electronics. [pdf]
L. Fan, C. Bai, Z. Pan and L. Chen "Towards Building the Semantic Map from a Monocular Camera with a Multi-task Network." [preprint]


3D Semantic Reconstruction from a Monocular Camera with a Novel Multi-task Network May 2018 - September 2018
We explore the interplay between low-level features for both depth and semantic prediction.
The proposed network can produce the depth and semantic maps simultaneously, which provides basic knowledge for further semantic map reconstruction.
We apply image segmentation techniques to refine the depth prediction to reduce the fluctuations caused by convolution layers.
The final map is saved in a memory-friendly way to present a large-scale urban scene.
The corresponding paper is recently submitted to the IEEE ICRA 2019. A video demo is uploaded to YouTube for demonstrating the result of our algorithm.

Using 3D Map Completion Method to Solve Ghosting Phenomenon
October 2017 - Now

The proposed method solves ghosting phenomenon caused by moving objects in a stereo-based 3D map.
The color and depth completion approach fills large area loss employing the planarity knowledge to propagate the structure.
The corresponding paper is accepted by IEEE Transactions on Industrial Electronics.

Planecell 3D Map Representation Method Developing
January 2017 - September 2017

The plancell extracts planarity from the depth-assisted image segmentation and then directly projects these depth planes into the 3D world.
The method demonstrates its advancement especially dealing with the large-scale structural environment, such as autonomous driving scene.
Intend to obtain instance-level segmentation result from semantic segmentation.
A video demo can be found at YouTube/youku.
The corresponding paper is under review.

DJI, Inc
Summer 2016

Visual Engineer Intern, Shenzhen, P.R. China
Developing 3D reconstruction and obstacle avoidance algorithms for the unmanned aerial vehicle based on the stereo camera.
Calibrating and rectificating of stereo fish-eye camera.
Developing stereo matching algorithms for fish-eye cameras which could give a broader range map.

Moving-object Detection Algorithm Developing
March 2016 - January 2017

The proposed method abandons the process of dense optical/scene flow calculation while giving pixel-level moving-object detection results. By accelerating on the GPU, it can run at 20 frames per second.
A video demo is uploaded to YouTube for demonstrating the result of our algorithm.

CNN-SPS Algorithm Developing
September 2015

Participating in the programming and paper writing.
The proposed method applies semi-global matching and slanted-plane model on the similarities from the CNN to produce accurate dense disparity maps.
The proposed method achieves the third place on the KITTI stereo 2015 benchmark in 2015.


Third Prize Merit-based Scholarship, SYSU 9/2014
Second Prize Merit-based Scholarship, SYSU 9/2015
Best Student Paper, IEEE Intelligent Vehicle Symposium 2018
First Prize Merit-based Grant, SYSU 9/2017
First Prize Merit-based Grant, SYSU 9/2018
National Merit Scholarship, SYSU, 9/2018

Technical Strengths

Computer Languages/Libraries/Frameworks C++, OpenCV, Python, Matlab, Tensorflow