UROP Proceedings 2020-21

School of Engineering Department of Computer Science and Engineering 99 AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: LIANG Shengnan / COMP Course: UROP1100, Spring This report gives an overview of the UROP project-AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data in spring semester 2021. First, it will introduce the project we are mainly focusing on (i.e. Kaggle contest: Indoor Location & Navigation-Identify the position of a smartphone in a shopping mall). Then introduce what relative knowledge was discovered and how to learn them from the perspective of our project and how to combine them with the project. Then two important steps in trajectory estimation will be introduced by comparing and analysis different approaches. In the end, this report will also give a summarize of the whole project, evaluate the performance of the project as well as state what have been learnt from the project and how these contribute to further personal development. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: SZE Kai Tik / COGBM Course: UROP2100, Spring This paper reviews the papers regarding density map application, especially for detection, aiming to find the challenges inside density map regression during the application in reality. In this research, difficulties about density map regression for detection will be concluded initially, and then the combination of detection use and counting use will be discussed later. After showing the challenges behind the density map regression for detection use, different neural network methods for detection will be reviewed and compared with each other so that this paper will help find the pros and cons among different training processes for reference of future research and study. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: ZHANG Weiwen / DSCT Course: UROP2100, Spring Crowd counting has been a hot topic with enormous potential in real-world application. This project is based on the approach that to train a CNN model to generate density maps with synthetic image datasets. Based on the previous work, we generate the dataset from Grand Theft Auto V (GTAV) in this paper by using GCCCL. Then we implement the domain adaptation to transfer the style of synthetic dataset to real world-world via CycleGAN. This project mainly generates synthetic dataset with GCC-CL and transfers the fake images by CycleGAN with SSIM embedding. And also tried to complete counting task on ShanghaiTech dataset with CSRNet. Source code of this report is available at: https://github.com/KevynUtopia/GTAV_MultiViewSyntheticDataset-DomainAdaptation

RkJQdWJsaXNoZXIy NDk5Njg=