School of Engineering Department of Computer Science and Engineering 83 VR Metaverse for Education Supervisor: BRAUD Tristan Camille / CSE Student: XU Borong / CPEG Course: UROP 1100, Spring This report focuses on exploring the potential applications of LoRA in stable diffusion. Through hands-on experience with LoRA, including model training, usage, and comparison of genAi results, we identify several advantages of the LoRA model. These include faster training speeds, greater freedom for result modification, and exciting outcomes from a module called “control net.” However, we also address issues such as overfitting during LoRA model training. In this report, we discuss the concept of LoRA in stable diffusion and present the experiment results. Additionally, we explore the potential use of LoRA in the field of VR metaverse. VR Metaverse for Education Supervisor: BRAUD Tristan Camille / CSE Student: YE Weicheng / COMP Course: UROP 1100, Fall This project mainly wants to integrate chatgpt and 3D model, and mainly wants to achieve two functions. The first one is to let the 3D model effectively communicate verbally and execute corresponding facial feedback, which is synchronization of mouth shape and voice, and also exhibiting different responses based on questions. The second is to achieve intelligent action-feedback. To specify, according to the input text, the 3D model will make reasonable action feedback. The general logic to achieve such a function is to use a large language model to analyze and make judgment, such that the 3D model will make the corresponding feedback and action, for example, input hello, the 3D model will lift up the hand to greet. AI + Healthcare: Research and Development of Intelligent Systems for Medical Diagnosis and Applications Supervisor: CHAN Gary Shueng Han / CSE Student: KWAN Kam To Christopher / CPEG Course: UROP 1000, Summer The integration of Sports AI into the healthcare sector is gaining momentum, offering users a range of exercise applications. However, many of these applications lack interactive features that provide personalized coaching and feedback. In the project, we aimed to develop the Sports AI application that provide interactive user experience by utilizing the machine learning and existing computer vision model. The current approach involves estimating the similarity score between a user's pose and the ideal pose using dynamic time warping (DTW) and L1 distance calculations. Furthermore, in this work, our objective is to enhance user experience by utilizing a pose analysis algorithm that extracts key points using the Yolov8 Pose estimation model and providing personalized pose suggestions by leveraging the Azure OpenAI chat completion API and prompt engineering techniques to interpret analysed pose data into human-like suggestions.
RkJQdWJsaXNoZXIy NDk5Njg=