Here is Xuanke Shi (史璇珂).
I am currently a Researcher at SenseTime Research. Previously, I received my M.Eng. degree from Zhejiang University in 2020, advised by R.A. Zhiwei Lin, Prof. Hongyao Shen, and Prof. Jianzhong Fu.
If you are interested in any aspect of my research, I would love to chat and collaborate. Please email me at Xuanke_Shi@zju.edu.cn.
My current research focuses on AIGC. I wish to devote my talent to this meaningful cause and bring well-being to society.
A unified multimodal generation framework that reformulates object detection, keypoint estimation, segmentation, depth estimation, surface-normal prediction, point-map generation, and camera-pose estimation within a unified multimodal model.
A unified framework for multimodal layout control and image composition, improving spatial consistency, controllability, and compositional quality in complex image-generation scenarios.
An AI-driven curling analysis and competition system developed for real-world Winter Olympics scenarios, integrating visual perception, multi-object tracking, and intelligent sports analytics.
Released Vision as Unified Multimodal Generation and the SenseNova-Vision model and corpus.
ConsistCompose was accepted by CVPR 2026.
Vimi, SenseTime's controllable character video generation model, was selected as one of WAIC 2024's “Eight Treasures”.
Contributed to SenseAvatar and developed its audio-driven talking-head generation algorithm.
Published the CurlingHunter system for the Beijing Winter Olympics.