UROP Proceedings 2020-21

School of Engineering Department of Computer Science and Engineering 119 Deep Video Super-Resolution Supervisor: CHEN Qifeng / CSE Student: ZHANG Juntao / SENG Course: UROP1100, Summer Text recognition is one of the earliest research fields in computer vision, aiming to enable computers to recognize text in various scenarios. As image information explodes on the internet, recovering text information from blurred pictures and providing readings for real-world automated machines become needed. Text super-resolution comes into rising as a vital branch. This report first summarizes the current practice, including related neural networks like CNN and RNN, image super-resolution method, and a scene text recognition method by Chen et al. (2021) Then give a research report focusing on text super-resolution on how to blur pictures in training can effectively train the super-resolution model. Lossless Point Cloud Compression Supervisor: CHENG Siu Wing / CSE Student: REN Zhengtong / COMP Course: UROP1100, Fall An effective, efficient, and lossless compression algorithm can facilitate the storage and transmission of large point clouds generated by modern 3D imaging devices. Current compression algorithms are based on octree, binary tree, clustering, etc. However, so far there is no attempt at using Steiner tree, a generalized version of minimum spanning tree, to solve this problem. In this UROP project, we first introduce a custom encoding method for floating-point numbers. Based on this encoding method, we then construct a rectilinear Steiner tree for the point cloud using the 1-Steiner heuristic. This progress report will describe the framework of our algorithm, highlight experimental results collected so far, and point out directions for future work. Automated Program Synthesis Supervisor: CHEUNG Shing Chi / CSE Student: CHEN Yijia / COSC Course: UROP1100, Spring Currently reporting HKUST Undergraduate Research Opportunity Program project on Automated Program Synthesis mainly focuses on polishing and extending an existing idea presented in the paper 'Programming by Example made Easy' to be submitted. Programming by Example (PBE) is one of the popular approaches to the long-lasting research problem of automated program synthesis, and with its potential further developed by the novel bi-directional search synthesis algorithm implemented in the PBE framework 'BEE' proposed by the paper, this UROP project studies some general knowledge about automated program synthesis, further compared the performance of the 'BEE' framework to some other methods, and partially extended the usage scenario of the 'BEE' framework to an unconventional area: Minecraft gaming.

RkJQdWJsaXNoZXIy NDk5Njg=