Shichao (Nicholas) Li
李世超

I am currently a research engineer at DJI automotive working on perception. I obtained my Ph.D. degree from the CSE department of HKUST, where I was advised by Prof. Kwang-Ting Cheng and have also worked with Prof. Chi-Keung Tang. Before that, I received my B. E. degree from ZJU.

My general research interests are computer vision and machine learning. Specifically, I'm interested in the representation, processing, and simulation of 3D data/models, its reconstruction/inference from a set of 2D projections, as well as its applications in autonomous perception and AR/VR.

Email  /  CV  /  Google Scholar  /  Linkedin  /  Github

The following is a list of my completed and on-going research projects in chronological order.

An extented study of Ego-Net, which presents the first binocular system for simultaneous outdoor 3D object detection and implicit shape estimation.

A novel multi-resolution approach for voxel-based stereo 3D object detection with instance-level modeling.

Ego-Net lets you recognize egocentric vehicle pose using a single RGB image.

This is the first study on utilizing synthetic data for 2D-to-3D human pose estimation.

GSNet reconstructs vehicle pose and shape jointly from a single RGB image.

How does a CNN make decision for classification/regression?

Numerical solution of Maxwell's equations for power transfer.

A finite difference algorithm solving drift-diffusion, current continuity and heat transfer equations jointly on a uniform 3D voxel grid.

Numerical simulation of acoustic wave.

The following is a list of my past projects.

I developed a real-time App on Andriod platform that can perform face detection, 3D faical landmark detection, head pose estimation and simple graphics rendering. The facial landmarks were detected by a cascade of random forest with pixel-difference feature. This project does not use deep learning. It was 100% C++/Java and 0% PyTorch. :)

This was an interferometry course project. A standard Michelson interferometer structure was adopted, light reflected from the unknown surface interferes with light reflected from a known surface.

By moving the known surface thus changing the light path, a sequence of interference pictures were captured. This process records the surface shape in the sequence. The surface shape was recovered in the demodulation process.

This was a simple micro-controller course project. Strings are made by laser diodes and photondetectors. Chord can be chosen by different buttons on this guitar. The system consists of C8051F020 (core), 74LS00 (logic), and LM386 (audio power amplifier). C8051F020 responds to user inputs and plays music stored in a SD card.

Lebensversicherung You can get the website template here from Jon Barron.