Qi Xu

Qi Xu 徐淇

My research focuses on multimodal large language models for video understanding and spatio-temporal reasoning.

I am a PhD student at Westlake University, advised by Peidong Liu. I work on multimodal data construction, SFT/RL post-training, and evidence-grounded evaluation, with applications in temporal grounding and embodied captioning.

Email GitHub CV Google Scholar

Research Experience

dots studio logo

rednote, dots Foundation Model Team

Research Intern

ByteDance logo

ByteDance, Data-AML-Doubao

Research Intern · Mentored by Ke Wang and Xiongcai Luo

  • Designed and scaled synthetic data pipelines for the embodied captioning and Doubao automotive projects.
  • Contributed to Doubao-Seed-Embodied-Caption model training.
TikTok logo

ByteDance, TikTok Global Live

Research Intern, LLM Strategy · Mentored by Xiangtai Li

  • Co-developed video MLLMs for temporal grounding, focusing on SFT/RL post-training, reward design, and evaluation.
  • Built multimodal training data and annotation pipelines for temporally grounded video understanding.

Education

Westlake University logo

Westlake University

PhD Student in Artificial Intelligence

Advisor: Peidong Liu

Wuhan University logo

Wuhan University

Master in Pattern Recognition and Intelligent Systems

Advisor: Prof. Shunping Ji

Wuhan University logo

Wuhan University

B.S. in Spatial Information and Digital Technology

Research direction

Understand. Reason. Learn.

Connecting vision, language, and 3D scene representations to improve spatial understanding and reasoning about the physical world.

Observe

Video Understanding

Temporal grounding and embodied captioning.

Interpret

Spatial Reasoning

Connecting visual evidence, language, and 3D scene representations.

Learn

Multimodal Post-Training

Data construction, SFT/RL, and evidence-grounded evaluation.

News

VideoZeroBench was released with code and data for evidence-grounded long-video evaluation.

Any 3D Scene is Worth 1K Tokens was released on arXiv.

Two advised-student papers were accepted to ECCV 2026.

Watch, Remember, Reason was released on arXiv.

HiCI was accepted to ICML 2026.

Towards One-to-Many Temporal Grounding was accepted to ICML 2026.

SIU3R received a NeurIPS 2025 Spotlight.

Selected Publications

Qi Xu is underlined. * Equal contribution. † Corresponding author.

SIU3R teaser
Perception NeurIPS 2025 Spotlight

SIU3R: Simultaneous Scene Understanding and 3D Reconstruction Beyond Feature Alignment

Qi Xu*, Dongxu Wei*, Lingzhe Zhao, Wenpu Li, Zhangchi Huang, Shunping Ji, Peidong Liu

Alignment-free framework for state-of-the-art 3D reconstruction and scene understanding from unposed images via a shared pixel-aligned 3D representation.

Project ArXiv Code
Watch, Remember, Reason teaser
Understanding arXiv 2026

Watch, Remember, Reason: Human-View Video Understanding with MLLMs

Jiahao Meng, Yue Tan, Qi Xu, Kuan Gao, Weisong Liu, Yanwei Li, Xiangtai Li, Lingdong Kong, Haochen Wang, Qianyu Zhou, Jiangning Zhang, Guangliang Cheng, Yunhai Tong, Lu Qi, Minghsuan Yang

A human-view survey organizing video MLLM research around watching, remembering, and reasoning for long, multimodal, and knowledge-intensive video understanding.

ArXiv GitHub
HiCI teaser
Understanding ICML 2026

HiCI: Hierarchical Construction-Integration for Long-Context Attention

Xiangyu Zeng, Qi Xu, Yunke Wang, Chang Xu

Efficient long-context modeling with hierarchical attention, extending LLaMA-2 to 100K tokens with only 5.5% additional parameters.

ArXiv Code
Perception arXiv 2026

Any 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale

Dongxu Wei*, Qi Xu*, Zhiqi Li, Hangning Zhou, Cong Qiu, Hailong Qin, Mu Yang, Zhaopeng Cui, Peidong Liu

Performs 3D scene generation directly within an implicit 3D latent space using 3D Diffusion Transformers.

Project ArXiv Code
ECCV 2026 Corresponding Author

SupIR-GS: Thermal Infrared Super-Resolution Novel View Synthesis with Imaging-Calibrated 3D Gaussian Splatting

Jin Liu, Haodong Li, Jiagang Chen, Dabin leng, Jiguang Li, Zhao Huang, Xiaoshuai Zhang, Qi Xu†, Zhiwen Zheng, Xingru Huang†

Reconstructs high-resolution 3D thermal scenes from low-resolution inputs via physics-informed degradation modeling.

ECCV
ECCV 2026 Corresponding Author

Physically Grounded Dual-Opacity Gaussian Splatting for Joint RGB-TIR Reconstruction

Jin Liu, Dabin leng, Jiagang Chen, Haodong Li, Jiguang Li, Zhao Huang, Xiaoshuai Zhang, Zhiwen Zheng, Xingru Huang†, Qi Xu†

A unified RGB-TIR reconstruction framework resolving cross-spectral visibility conflicts through thermal field modeling.

ECCV
E-MoFlow teaser
Perception NeurIPS 2025

E-MoFlow: Learning Egomotion and Optical Flow from Event Data via Implicit Regularization

Wenpu Li, Bangyan Liao, Yi Zhou, Qi Xu, Pian Wan, Peidong Liu

Unsupervised framework for joint ego-motion and optical flow estimation using implicit neural representations and geometric constraints.

Project ArXiv Code

Contact

I am open to research conversations and collaborations on multimodal large language models, video understanding, spatio-temporal reasoning, and multimodal data and post-training.

insomniaaac@qq.com github.com/insomniaaac Google Scholar