【軟工書報討論】6月4日（三）Shuai, Hong-Han (帥宏翰) (Professor and Associate Chair, Department of Electronics and Electrical Engineering Director, Institute of Electrical and Computer Engineering National Yang Ming Chiao Tung University)
2025-06-02 14:49:54
演講時間: 114年6月4日（三） 14:00~16:00
演講地點: E6-A203教室
演講者: Shuai, Hong-Han (帥宏翰)
(Professor and Associate Chair, Department of Electronics and Electrical Engineering Director, Institute of Electrical and Computer Engineering
National Yang Ming Chiao Tung University)
講題：Advances in Controllable Diffusion for Multimodal Generation
大網: In this talk, I will present recent advancements from our lab at NYCU on generative modeling with diffusion frameworks, targeting precise image and scene synthesis. We address key limitations in prompt interpretation, fine-grained image harmonization, and SAR recognition through innovations that eliminate training and prompt dependencies, integrate LLM-guided semantics, and introduce tailored pretraining strategies. Featured works include our training- and prompt-free painterly harmonization method (AAAI 2025) and scene generation with LLM-assisted prompt understanding (ACM MM 2024). These contributions push the boundaries of controllability and generalization in multimodal generation, with broad implications across vision, language, and remote sensing domains.
