Prof. K.H. Wong's M.Sc research projects  (CMSC5720/1)

  K.H. Wong,  updated 2026.4.4

M.Sc. projects to be supervised by Prof. Kin Hong Wong,  year 2026-7
 
Project 1: 3D rendering development for virtual reality and game developers
We would like to develop a mobile app based on the method of Gaussian Splatting for real-time 3D reconstruction. Users can capture images or videos, converting them into high-quality 3D models with advanced photo-realistic rendering in real time. Such tools should have significant commercialization potential in gaming and Virtual reality. You may see the live demo  at https://poly.cam/tools/gaussian-splatting  or get the base code from  https://github.com/graphdeco-inria/gaussian-splatting .
 
Proect 2: AI guide dog for the visually impaired
In this project we investigate the development of an AI-powered guide robot/dog to assist visually impaired users in navigating indoor and outdoor environments. Using computer vision and embedded AI, the robot will analyze the surrounding images make decision for safe navigation. We may consider using mobile edge devices, such as NVIDIA Jetson or Raspberry Pi, which enable low power consumption and enhance mobility.

Project 3. Neural network without backpropagation
A new paradigm in neural network training has recently emerged: NoProp, which eliminates the need for traditional backpropagation. This approach dramatically reduces memory consumption—by as much as 70%—and offers a glimpse into what future machine learning methods may look like. Students are encouraged to explore and experiment with this model, investigating ways to refine and extend it for even greater performance and efficiency.
Reference:
NoProp: Training Neural Networks without Full Back-propagation or Full Forward-propagation Qinyu Li, Yee Whye Teh, Razvan Pascanu
https://arxiv.org/abs/2503.24322

Project 4: Free AI project using Large Language Model (LLM) you may propose a project you find interesting that uses LLM for practical applications.

Project 5: The AI Singalong Machine: From Solo Singing to a Full Band. In this project, you will build an AI Singalong Machine using Python, and you do not need any music theory knowledge to do it! First, your computer will listen to a recording of a person singing and automatically convert their voice into digital musical notes. Next, a smart AI model will read these notes, analyzing the patterns purely as data to generate matching background music—such as piano or drums—that fits the voice perfectly and generate the mp3 sound file. It is a fun way to learn AI methods such as music signal processing,  neural networks and transformers etc.











Guidelines for students  in CMSC5720/1
Normally,  meet the supervisor weekly (or biweekly) face to face or by Internet-Zoom, the Zoom link can be found in the course page at https://blackboard.cuhk.edu.hk .
Send 1-2 page weekly report (through the "Weekly report assignment" in the course content of the course page at  https://blackboard.cuhk.edu.hk)  to the supervisor to discuss your ideas and work achieved.
---------------------------------------------------------------------------
CMSC7250 (Term1): total 13 weeks
4 weeks: Define details of the project and write plans for the rest of the term.
4 weeks: Literature search and testing of open source programs related to the work.
4 weeks: Develop your own program, can be an integration of open source routines/libraries from others.
1 week:  Report writing and presentation preparation.
-----------------------------------------------------------------------
CMSC7251 (Term2): total 13 weeks
3 weeks: Improve work in the first term. Add original features that different from others to the project.
3 Weeks: Enhance efficiency, add extra capabilities and features to the project.
4 Weeks: Testing of the system and analysis of data. Increase robustness and accuracy. 
3 Weeks: Final Report writing, presentation preparation and rounding up.

Research and Thesis Writing























------------   previous projects -------------------------
2023-24
MSC projects to be supervised by Prof. Kin Hong Wong,  year 2023-24,         2023.6.2

Artificial Intelligence (AI) projects: The purpose of these projects is to train students to learn about AI theories and programming, it may involve the use of tools such as Tensor-flow, Keras, Generative Pre-trained Transformer (GPT) etc. Students are free to choose to work on one of the following topics and applications.
1)    Detect the type of sea vessel (ships) from pictures taken at 300m-1000m away. We need to measure the travel speed and location of the sea vessels. This industrial project may be supported by a science park company which can provide us with the data set and computation power.
2)    Extraction of vital information from sounds and images from videos. Using modern AI methods, we may be able to extract useful information from video sources. It is an industrial project to extract useful information (product type, weights, and code) from working videos taken by mobile phones. A sample input video can be found at (http://www.cse.cuhk.edu.hk/~khwong/www2/cmsc5720/fish_market.mp4). 
 3)    Free projects: students can propose projects that involve modern AI techniques. Such as experiments with Large Language Model (LLM) and Generative Pre-trained Transformer (GPT) .
Details of the projects can be found at http://www.cse.cuhk.edu.hk/~khwong/www2/cmsc5720/cmsc5720.html .

2022-23

MSC projects to be supervised by Prof. Kin Hong Wong  2022-23,                  2022.5.30

a)      Computer vision processing research:

i)        Generate a sentence from an image: https://www.youtube.com/watch?v=c_bVBYxX5EU .  The idea is to tell a story from a picture. Human can do it effortless, now Artificial Intelligence may be applied to solve the problem. We have done some primary work and the sentences produced are accurate but small and fragmented. We would like to improve the performance in the coming term. It can be applied to assist visually impaired persons to read pictures or let them to better understand the surrounds.

i)        Vision based gesture or sign language recognition: This is useful to automatically recognize hand gesture languages to assist communication between a normal person and those lost the ability to speak. This can also be used to develop exercise tutorial systems for training tutors. Demos https://www.youtube.com/watch?v=vTC0QKR_uM0 or https://www.youtube.com/watch?v=doDUihpj6ro .

 

b)      Computer audio processing research:

i)        Voice cloning and applications: https://www.youtube.com/watch?v=1WN8Jhfd4uM  This interesting demo may inspire new applications of audio sound generation for the music industry.

ii)      Music genre classification: https://www.youtube.com/watch?v=szyGiObZymo   This is useful for the online music providers to recommend suitable music to users.

iii)    Audio synthesis and tone changes: https://anonymous84654.github.io/RAVE_anonymous/ Many non-native English speakers speak English with a local accent. The idea is to turn the non-native English recording into a speech as if it is spoken by a native speaker. The idea can be applied to all different target and destination languages, which is useful for students and travelers.





  2021-22
2020-21 MSC_projects for CMCS5720(term1) and CMCS5721(term2) :
2018-9
   2017-8
   2016-7

  2015-6

2014-5
idea1