Title of the Talk: Harnessing Generative Priors for Image Super-Resolution
Speaker: Chen Change Loy, Nanyang Technological University, Singapore
Abstract: In this talk, I will present an overview of recent advancements in harnessing generative priors for image super-resolution, particularly focusing on blind face restoration and real-world applications. We discuss how leveraging learned discrete codebook priors, and robust diffusion priors can significantly improve restoration quality and robustness under complex and unknown degradations. In particular, I will discuss CodeFormer, a Transformer-based prediction network that uses a codebook prior to tackling the inherent uncertainty in blind face restoration, and the DifFace, an approach that exploits a pre-trained diffusion model to cope with unseen degradations without complicated loss designs.
Title of the Talk: Unlocking the Secrets of Human Movement: A Look at the Cutting-Edge Progress in Gait Recognition
Speaker: Yasushi Yagi, Osaka University, Japan
Abstract: The way we walk, or our gait, can reveal a lot about our personal information, such as our identity, gender, age, emotions, intentions, and even our health. This has led to extensive research on gait recognition not just in biometrics, but also in fields like biomechatronics and healthcare. However, applying gait recognition to real-world problems presents a challenge due to various fluctuating factors that need to be taken into account. In this presentation, I will discuss recent progress in gait recognition research by our group, which focuses on the diversity of observations, including cross-view and occlusion, as well as the uncertainty of annotation. Additionally, I will touch on video-based gait analysis, which can play a critical role in medical and wellness fields if time permits.
Title of the Talk: The Art of Non-Verbal Communication: Decoding Facial Expressions and Gestures for Seamless Human-Robot Interaction
Speaker: Xilin Chen, Chinese Academy of Sciences, China
Abstract: As robots become more prevalent in our daily lives, their ability to understand and respond to human communication is becoming increasingly important. While natural language processing (NLP) has enabled robots to capture human intentions from verbal communication, understanding non-verbal cues such as facial expressions and gestures remains a challenge. However, these non-verbal channels are crucial for conveying user intention to robots, especially for human-centered robotics. In this talk, I will explore the roles of non-verbal channels in human communication, which provide additional information to emphasize key points and express attitudes. I will also discuss recent efforts in understanding facial expressions and gestures, both conscious and unconscious, to improve the interaction efficiency between humans and robots.