主讲人
Xudong Mao Sun Yat-sen University 时间 2026年6月25日 星期四 上午 10:00-11:00 地点 学院104会议室
Abstract Controllable visual generation aims to create and manipulate visual content in precise alignment with human intent. Despite remarkable advances in deep generative models, achieving reliable, fine-grained, and semantically faithful control remains a fundamental challenge. This talk presents our recent efforts toward controllable visual generation, focusing on two closely related directions: personalization and image editing. The first part addresses personalized text-to-image generation, exploring how user-provided visual concepts can be more effectively integrated into generative models. The second part introduces image editing frameworks that learn editing semantics from visual examples or natural-language instructions, enabling models to transfer complex edits while maintaining robustness throughout iterative editing processes. Biography Dr. Xudong Mao is an associate professor at the School of Artificial Intelligence, Sun Yat-sen University. He received his Ph.D. from City University of Hong Kong and his B.Eng. from Nankai University. His research focuses on diffusion models, generative adversarial networks, and deep generative models. Dr. Mao has published more than 40 papers in major academic conferences and journals, with over 9,000 citations on Google Scholar. He has served as an Area Chair for conferences such as NeurIPS and ICME, and has been recognized on Stanford University’s list of the World’s Top 2% Scientists.
