Siebel School Speaker Series Master Calendar

Abstract: Although Large Language Models are originally designed for specific tasks, like translation, it now serves foundation models for many different complex reasoning, autonomous agents, and multimodal problem-solving. Similarly, generative vision models have captured deep priors of the visual world and should generalize beyond the synthesis of high-fidelity visual content. In this talk, we share our recent efforts in extending these models into functional domains.

We first discuss our work on leveraging pretrained models to enhance image and video processing. This includes FlashVSR for real-time diffusion-based super-resolution, UltraFusion for high-dynamic imaging, CubeComposer for 4K 360° video synthesis, and InstantRetouch for efficient, instruction-guided editing. Furthermore, we explore how video generative models facilitate advanced 3D and 4D generation. This is demonstrated by 4DSloMo, which reconstructs high-speed scenes from asynchronous captures, and AnyRecon, which enables scalable 3D reconstruction from arbitrary sparse inputs. Finally, we briefly explore how generative models may learn to simulate physical laws.

Speaker Bio.: Prof. Tianfan Xue (https://tianfan.info/) is an Assistant Professor at the Multimedia Lab (mmlab) in the Department of Information Engineering at the Chinese University of Hong Kong. Prior to this, he worked in the Computational Photography Team at Google Research for over five years. He received his Ph.D. degree from the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT) in 2017. He also holds an M.Phil. degree from CUHK, obtained in 2011, and a Bachelor’s degree from Tsinghua University. His research focuses on computational photography, 3D reconstruction, and generation. The anti-reflection technology he investigated is utilized by Google Photoscan, which boasts over 10 million users. His recent work on bilateral based 3D reconstruction has won SIGGRAPH Honorable mention 2024, work on HDR fusion won the CVPR Best Demo Honorable Mention. He also served as an area chair for WACV, CVPR, NeurIPS and ACM MM.