School of Engineering Department of Computer Science and Engineering 104 Video Analytics and RF People Sensing for Smart City and New Retail Supervisor: CHAN Gary Shueng Han / CSE Student: TANG Tianhao / COSC Course: UROP3100, Fall In this semester, I have learnt about graph neural networks (GNN) and its applications. I first got to know what a graph neural network is, what it is different and similar to convolutional neural networks, and how it works. Then I learnt how to build a graph neural network based on the framework. Furthermore, I have read some papers that focus on extensions of GNN like the heterogeneous graph. Besides, I have also participated in a project that is going to build a model to classify the spam users from the review data from 58, a famous site that provide services on lives. This report will state them in detail. Video Analytics and RF People Sensing for Smart City and New Retail Supervisor: CHAN Gary Shueng Han / CSE Student: TANG Tianhao / COSC Course: UROP4100, Spring In this semester, I continuously focus on dyslexia or dysgraphia project. This semester, we mainly focus on using computer vision and machine learning technologies to identify the faults and problems in writing characters, like missing components or flipped the character. To achieve this, we first tried building a system to distinguish the chirality of writing characters globally or partially. We then tried several ways to develop a system that can decompose the character components and strokes and giving out problems targeting each part of the characters. This report will give details on what we have done and what we will do in the future. Video Analytics and RF People Sensing for Smart City and New Retail Supervisor: CHAN Gary Shueng Han / CSE Student: CHANG Hong-yuan / DSCT Course: UROP1000, Summer Videos captured by cellphones or surveillance cameras often suffered from blurriness, which motivates research interests in video super-resolution (VSR). The primary goal of VSR is to computationally generate high-resolution videos given corresponding low-resolution inputs. In this UROP project, we specifically interest in large scaling factors VSR for real-world video clips. We first briefly discuss the problem setting of super-resolution (SR). Next, we introduce three state-of-the-art models for SR problems and conduct several experiments. Finally, we analyze the pros and cons of different models qualitatively and suggest possible directions for future improvement.