| Author: | Liu, Yujun |
| Title: | Mathematical modeling and advanced diffusion techniques for age-specific facial image generation |
| Advisors: | Lam, K. M. Kenneth (EEE) |
| Degree: | M.Sc. |
| Year: | 2023 |
| Pages: | xi, 121 pages : color illustrations |
| Language: | English |
| Abstract: | In the past few years, machine learning has made huge strides, mainly because of better hardware and smarter deep learning techniques. This paper focuses on improving Denoising Diffusion Probabilistic Models (DDPMs), a key area in this growth. We’re particularly interested in the early stages of stable diffusion. Our work introduces ’rectified flow,’ a new method using Ordinary Differential Equation (ODE) models to improve the denoising steps in the latent space, particularly from the 30th to 50th iterations. The ’rectified flow’ method is about blending theory with practical machine learning tasks. It’s a smoother approach compared to the usual step-by-step methods. Based on recent machine learning advancements, it allows for efficient and high-quality image creation in just one step. This is a big leap forward for tasks like image editing, where we don’t have pre-set examples. Our tests show that this method outperforms the current best techniques, setting a new standard for generative models. The new method of the SD-RegLoRA RecFlow technique in our study is not just about fusing facial features or age elements; it’s about opening a whole new range of possibilities. It’s particularly useful for industries like gaming and art, where AI-driven workflows are common. This technique could significantly shorten project timelines and ensure high-quality, style-consistent outputs. As we look to the future, it’s clear that continuous research and innovation are key to unlocking the full potential of machine learning technologies. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 8268.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 17.02 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13817