School of Engineering Department of Computer Science and Engineering 100 Deep Video Super-resolution Supervisor: CHEN Qifeng / CSE Student: MA Mingfei / CPEG Course: UROP 1100, Fall Large text-to-image models achieved a remarkable leap in the evolution of AI, enabling high-quality and diverse synthesis of images from a given text prompt. However, these models lack the ability to mimic the appearance of subjects in a given reference set and synthesize novel renditions of them in different contexts. And the newly developed technique Dreambooth presents a new approach for "personalization" of text-toimage diffusion models. Given as input just a few images of a subject, the technique can capture it and then synthesize the subject in diverse scenes, poses, views, and lighting conditions that do not appear in the reference images. The technique is called Dreambooth because it’s like a photo booth, but once the subject is captured, it can be synthesized wherever your dreams take you. It’s a powerful tool for personalizing textto-image diffusion models. Deep Video Super-resolution Supervisor: CHEN Qifeng / CSE Student: WANG Yifan / DSCT Course: UROP 2100, Fall During this fall semester, I participated in the UROP2100 program supervised by Professor Chen Qifeng. Since I paid more attention to video-generating and video-editing in the summer semester, Professor Chen asked me to work with his graduate Ph.D. student, Professor Lei Chenyang, now in CAIS, HK. Under his guidance, I focused on two projects—All-in-one-Deflicker and CoDeF: Content Deformation Fields for Temporally Consistent Video Processing. Both projects emphasize the time consistency of a video and have some similarities in the idea. In this report, I will show my progress this semester, but some outcomes may not be shown because there is something wrong with my remote server after Nov 21st. Generative AI Supervisor: CHEN Qifeng / CSE Student: CHEN Chenle / COMP Course: UROP 1100, Spring This project investigates the use of generative AI in producing short videos from text descriptions. By leveraging AI tools such as Midjourney for image creation and DynamiCrafter for animation, we developed dynamic visual content. The final video was assembled using Capcut, integrating AI-generated dubbing, background music, and sound effects. The project highlights the potential of AI in creative video production, focusing on a children's animation titled "Digital Honeycomb", which centers on environmental awareness and digital innovation. Despite the impressive visual outputs, significant human oversight was necessary to refine the AI-generated content, pointing to current limitations in AI's narrative understanding. The project suggests future exploration into more integrated AI solutions for streamlined content creation.
RkJQdWJsaXNoZXIy NDk5Njg=