UROP Proceedings 2021-22

School of Engineering Department of Computer Science and Engineering 108 Deep Video Super-resolution Supervisor: CHEN Qifeng / CSE Student: ZHANG Juntao / COSC Course: UROP2100, Summer Scene text image super-resolution aims to enhance the readability of low-resolution scene text images by increasing their original resolution. It can be used as a pre-processing step in scene text image recognition or to assist human readers. By learning end-to-end using paired high-resolution (HR)-low-resolution (LR) images, a scene text super-resolution model can significantly improve the accuracy of text recognition models of lowquality images and super-resolve human-unreadable blurred document images. In this paper, we proposed a Prior-Guided Super-Resolution (PGSR) structure to address this problem, which was inspired by the Text Attention network (TATT) . Furthermore, we proposed an instance of PGSR, hybridizing two state-of-the-art models, TATT and Scene Text Telescope (STT). To test the performance, we evaluate the output by traditional image super-resolution PSNR/SSIM metrics, as well as the recognition accuracy of text recognition models like CRNN. Code is available at GitHub. Neural Rendering Supervisor: CHEN Qifeng / CSE Student: CHENG Yize / COSC Course: UROP1100, Summer Autonomous Driving (AD) is an emerging field that is growing rapidly. This has been particularly the case since the development of deep learning architectures, such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and Graph Neural Networks (GNN). Common tasks for autonomous driving applications include object detection and tracking in videos, motion trajectory prediction and trajectory planning. In this UROP project, we studied the paper Learning Lane Graph Representations for Motion Forecasting. The proposed LaneGCN architecture for motion forecasting is still very commonly used in trajectory prediction tasks in current autonomous driving applications. In this report, we try to unscramble this paper in our words based on our own understanding and make the ideas more comprehensible for beginners than the presentation in the original paper .