 |
Si Lu |
Ph.D.
Department of Computer
Science Maseeh College
of Engineering and Computer
Science
P.O. Box 751
Portland State
University
Portland, OR 97207-0751
Email: lusi "at" pdx.edu
Resume |
| |
| I'm
a research assistant at Computer Graphics
& Vision Lab. Graduating soon and seeking for full-time positions in
Computer Vision, Machine Learning/Deep Learning as well as Computational
Photography. My advisor is Prof.
Feng Liu. I received my B.E. and M.S. degree from Tsinghua University. More
than 9 years' experience on Computer Vision and Graphics topics such as
novel view synthesis, video stabilization, image stitching, object detection
and computational photography such as image denoising and inpainting.
Passionate about developing and implementing novel computer vision
algorithms. Skilled in implementing state-of-the-art computer vision/image
processing algorithms. Skilled programmer with C/C+ +/Matlab/OpenCV/OpenGL.
Familiar with CNN/ResNet/DenseNet/Faster RCNN/Mask RCNN. |
| |
|
Publication |
Si
Lu, Xiaofeng Ren, and Feng Liu. Depth Enhancement via Low-rank Matrix Completion.
IEEE CVPR 2014, Columbus, OH, June 2014. Project
website. Poster. |
| |
|
Experience |
|
Computer Vision Intern - Lytro Immerge (VR) |
Moutain View, CA, Jun., 2017 - Sep., 2017 |
|
 |
Multi-view 3D reconstruction ( C/C++/OpenCV )
• Processing HD (2048×2048) RGB-D images
• Signigicant fine detal improvements for
VR depth maps (left)
• Depth enhancement for light filed
(multi-view) camera arrays (middle)
•
Depth enchancement for stiched 360-degree videos in Virtual Reality (right)
•
Integrated into Lytro Immerge (VR) production pipeline
for depth |
 |
 
 |
|
|
|
Projects |
 |
Highspeed Videos from Camera Arrays (2016-2017)
•
Develope an algorithm to generate highspeed videos from camera arrays
•
Use novel views synthesis to do frame interpolation among cameras
•
Use image-based rendering techniques, e.g., feature matching/warping
•
Plausible and parallax-free novel views can be robustly interpolated
•
Works for challenging scenes with large camera and object motions
•
Submitted to CVPR 2018 (Video
demo) |
|
|
|
  |
Patch Matching for Image Denoising (2017-present)
Develope a clustering-based approach to consistently improve patch-based
denoising techniques' (like BM3D) performance by using learned patch
descriptors trained via deep learning. (in progress, submitting to
ECCV 2018) |
|
|
|
  |
Depth Enhancement (2012-2014)
Develop a depth map
enhancement algorithm that performs depth map completion and denoising
simultaneously for RGBD cameras like Microsoft Kinect and Xtions Pro.
Project
website.
Poster. |
|
|
|
 |
Flash Light Detection for Unmanned Aerial
Vehicles (2015)
Developed an algorithm that allows real-time
flash light detection for Unmanned Aerial Vehicles. By designing specific
color-based features, we achieved a 89.2% detection rate with a small false
positive rate (0.6%). |
|
|
 |
Robocup Team Leader (2007-2012)
Project website.
•
Developed a robot vision system with accurate and real-time (20 FPS) objects
(football, field lines and goals) detection and ball-locating/self-locating
(C++/Windows Mobile SDK/Linux).
• Developed a
wireless vision parameter tuning interface to allow fast camera calibration
before real games. By utilizing this interface, I reduced the parameter
tuning time by more than 70% before each game (Matlab, C). |
|
|
|
|
Teaching |
|
CS 447/547-Computer Graphics (instructor): teach
Computer Graphics topics on movies, games, animations and 3d rendering.
Helped students with two course projects - a mini-Photoshop (FLTK/OpenCV)
with basic image processing and an Amusement Park 3D animation rendering
project (OpenGL). |
|
CS
410/510-Computational Photography (instructor): teach
research topics ranging from concepts of digital camera and photography to
computer vision/graphics techniques, including high dynamic range imaging,
panorama stitching, image segmentation & matting, video stabilization,
virtual reality basics, deep learning in computer vision etc.. |
|
