Xuhan Sheng
Ph.D. student in Artificial Intelligence, Institute of Science Tokyo.
Research background in diffusion models, memory-efficient training and inference, vision-language models, and video data pipelines. Co-first author of an ECCV 2024 Oral paper.
Research interests: video generation, AI agents, and large language models.
Seeking: Long-term research internships; available to start as soon as arrangements permit. Full-time availability subject to local work authorization and company policy.
Education
Institute of Science Tokyo, Department of Computer Science
Oct. 2026 – PresentPh.D. student in Artificial Intelligence | Advisor: Prof. Rio Yokota
Earliest graduation: Sept. 2029; option to graduate in Apr. 2030.
Peking University, Computer Application Technology
Sept. 2023 – Jun. 2026Master's degree | GPA: 3.87/4.00 | Advisor: Prof. Jian Zhang
Dalian University of Technology, Artificial Intelligence
Sept. 2019 – Jun. 2023Bachelor's degree | GPA: 4.17/5.00 | Advisor: Prof. Xu Jia
Research and Industry Experience
Xiaomi Auto, Shanghai Team — World Model Research Intern
Aug. 2026 – Present- Contributing to Xiaomi Auto World Model 2.0 (forthcoming), with primary responsibility for video super-resolution.
Tencent IEG, Game AI Engine Department — Algorithm Intern
Mar. 2026 – Jun. 2026- Developed Qwen-Image-based super-resolution for game visuals and production assets.
- Built data-pipeline operators for video generation, including streaming shot segmentation that converts long videos into single-shot clips; work was integrated into the production pipeline.
OPPO Research Institute — Imaging Algorithm Engineer Intern
May 2025 – Sept. 2025- Researched vision-language models for locating and annotating perceptual artifacts in AI-generated images to support image enhancement. AI Talent Program; advisor: Prof. Lei Zhang.
Research and Publications
* Equal contribution. Full publication list.
OmniSSR: Zero-shot Omnidirectional Image Super-Resolution using Stable Diffusion Model
- Contributed to method design, implementation, and evaluation of zero-shot panoramic super-resolution using diffusion priors, tangent-view information exchange, and gradient decomposition.
PanoForcing: Enforcing Memory-Scalable Full-Context for Blind Panoramic Super-Resolution
- Led method design, implementation, core experiments, and manuscript writing. Combined selective autograd subgraph routing with Fourier pseudoinverse guidance for memory-scalable, full-context one-step panoramic super-resolution.
RealOSR: Latent Guidance Boosts Diffusion-based Real-world Omnidirectional Image Super-Resolutions
- Major contributor to implementation, experiments, and manuscript writing for one-step diffusion-based panoramic restoration with latent-space guidance.
OmniDrag: Enabling Motion Control for Omnidirectional Image-to-Video Generation
- Contributed to the Move360 large-motion dataset for controllable panoramic image-to-video generation.
Label-guided Facial Retouching Reversion
- Developed hierarchical adaptive instance normalization to mitigate color drift in facial retouching reversion.
Bridging the Point to Boundary Gap for Point-supervised Temporal Action Localization with Single-stage Inference
- A hierarchical framework converting sparse point annotations into boundary-accurate pseudo labels via Gaussian-prior boundary enhancement.
- Assisted in manuscript writing and submission.
OPDN: Omnidirectional Position-Aware Deformable Network for Omnidirectional Image Super-Resolution
- Contributed a spatial-frequency fusion module and fisheye data collection to the NTIRE 2023 omnidirectional SR challenge winning team.
Skills and Recognition
Programming: Python, PyTorch.
English: IELTS Academic 7.5/9.0.
- NTIRE 2023 omnidirectional SR challenge — winning team.
- Peking University Academic Excellence Award — 2024.
- Dalian University of Technology University-level Outstanding Graduate — 2023.
- Dalian University of Technology First-class Academic Excellence Scholarship — 2022 and 2020.
- Dalian University of Technology Second-class Academic Excellence Scholarship — 2021.