Layer 4

Layer 5

Layer 6

Layer 7

Net Output



Explain ⇓

Live Demo: Sketches to Photos

The live demo utilizes a Neural Network to synthesis photos with face sketches, which is helpful for various applications, e.g., identifying the suspect. The Neural Network in this example is regressing pixel values live in your browser based on ConvNetJS, a JavaScript based ConvNet library. It takes a pixel of a sketch and transforms it through a series of functions into RGB values as the output. The transformed representations in this visualization can be losely thought of as the activations of the neurons along the way. The Sketch-Photo pairs are from the CUHK Face Sketch Database (CUFS). By the end of the class, you will know exactly what all these numbers mean.


Teaching Assistants


[Mar 7] On Mar 9th, 16:30-17:30, we will be using the ZOOM address to conduct quiz 1. If you are just sitting in the class, please join the ZOOM session at 17:40 on Mar. 9th.
[Feb 23] The slide for `Network Architectures for Image Understanding` is updated.
[Feb 21] Three new lecture slides have been uploaded.
[Feb 3] The next tutorial will last for 1.5hrs and will be held on Feb 4. This tutorial will mainly focus on the basic usage of Pytorch and its debugging.
[Jan 15] Video recordings for week 1 have been uploaded. Please check the lecture page.
[Jan 11] This course would be online until further notice. Please use the ZOOM link for attending lectures and tutorials.
[Jan 4] Welcome to ELEG 5491 Introduction to Deep Learning!

Course Description

This course provides an introduction to deep learning. Students taking this course will learn the theories, models, algorithms, implementation and recent progress of deep learning, and obtain empirical experience on training deep neural networks. The course starts with machine learning basics and some classical deep models, followed by optimization techniques for training deep neural networks, implementation of large-scale deep learning, multi-task deep learning, transferred deep learning, recurrent neural networks, applications of deep learning to computer vision and speech recognition, and understanding why deep learning works. The students are expected to have some basic background knowledge on calculus, linear algebra, probability, statistics and random process as a prerequisite. The course offered in Spring 2021 features:

Time and Venue

Term 2 (January - April), 2021
Zoom Link:
  • Tuesday, 16:30-18:15
  • Wednesday, 15:30-16:15
  • Thursday, 14:30-15:15 (every other week)
    starting from 21/1/2021

Contact information

Hongsheng LI:
Xiaoyang GUO:
Peng GAO:


Ian Goodfellow and Yoshua Bengio and Aaron Courville, “Deep Learning,” MIT Press, 2016

Grading Policy

Assignments: 30%
Quiz: 30%
Final Project: 40%


I am a student outside the EE department, can I register in the class?
Yes, you are welcome to register. For graduate students outside our EE department, you can fill in this form and ask approval from both your supervisor and the course instructor during the add/drop period.
Is this course hard for undergrad students?
The course is designed for senior undergrad and graduate students. It is not for the faint of heart. However, we will show lots of interesting cases and hands-on experience about deep learning models. Some part of the lectures requite calculus and linear algebra, but we will walk you through those knowledge. We think for undergrads, you will learn a lot at the end of the course through lectures, tutorials and the final project.
Can I work in groups for the Final Project?
No. The final project is done individually and details will be announced later.


We also provide the 2019 lecture notes and tutorials..

Past Contributors

We sincerely thank all the contributors who made great efforts in supporting this course:
Prof. Wanli Ouyang, Prof. Hongsheng Li
Dr. Xingyu Zeng, Dr. Zhe Wang, Dr. Tong Xiao, Dr. Xiao Chu, Dr. Wei Yang, Dr. Kai Kang, Dr. Hongyang Li