Multimodal AI + Imaging Collaboration

Xiangming Wang builds multimodal models for robust visual understanding and restoration.

Ph.D. candidate at HIT Shenzhen, focusing on vision-language modeling, multi-sensor fusion, and optimization-driven learning.

My work connects multimodal representation learning with practical deployment, from vision-language guided restoration to multi-sensor RAW imaging pipelines. I focus on methods that are structurally faithful, degradation-aware, and efficient enough for real-world devices.

Start Collaboration Download CV

About

Profile

I am from the Department of Computer Science at Harbin Institute of Technology (Shenzhen). My recent research increasingly centers on multimodal modeling for image restoration, including vision-language guidance and multi-sensor fusion. I build optimization-aware methods that preserve structure under complex degradations while remaining practical for deployment.

Research Focus

Topics

Vision-Language Restoration

Vision-language guided deep unfolding and generative restoration for degradation-aware recovery.

Multimodal Sensing

Multi-sensor RAW and hyperspectral reconstruction with cross-modal consistency constraints.

Generative + Optimization

Autoregressive/diffusion modeling integrated with convex and non-convex optimization principles.

Education

Timeline
Ph.D. Candidate in Computer Science, HIT Shenzhen Sep 2024 - Present

Expected 2029 · GPA: 89.942/100 · Rank: 4%.

B.Eng in Computer Science, HIT Shenzhen Sep 2020 - Jun 2024

GPA: 89.576/100 · Rank: 19%.

Experience

Timeline
Algorithm Intern, Y-Lab, OPPO Research Institute Sep 2025 - Present

Unified Denoising & Demosaicing for Mobile Imaging Pipeline with multimodal sensor adaptation.

  • Designed a joint optimization framework for RAW-domain denoising and demosaicing in mobile ISP pipelines.
  • Investigated adaptation strategies for heterogeneous multi-sensor data to enable robust multimodal fusion.
  • Prototyped lightweight deployment methods for multimodal inference under strict on-device latency constraints.

Technical Skills

Stack

Modeling

Vision-language modeling, deep unfolding, diffusion/autoregressive generative models, multimodal fusion.

Data

Synthetic data generation, degradation modeling, cross-modal alignment, spatiotemporal consistency metrics.

Engineering

PyTorch, mobile ISP pipeline development, multimodal training pipelines, CUDA acceleration basics.

Publications

Selected

  • Vision-Language Gradient Descent-driven All-in-One Deep Unfolding Networks

    Haijin Zeng*, Xiangming Wang* (Co-first author), Yongyong Chen, Jingyong Su, Jie Liu

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    Key contribution: Vision-language guidance for degradation-aware restoration with preserved structural fidelity.

  • Deep LoRA-Unfolding Networks for Image Restoration

    Xiangming Wang, Haijin Zeng, Benteng Sun, Jiezhang Cao, Kai Zhang, Qiangqiang Shen, Yongyong Chen

    IEEE Transactions on Image Processing, 2026

  • OTLRM: Orthogonal Learning-based Low-Rank Metric for Multi-Dimensional Inverse Problems

    Xiangming Wang, Haijin Zeng, Jiaoyang Chen, Sheng Liu, Yongyong Chen, Guoqing Chao

    Proceedings of the AAAI Conference on Artificial Intelligence, 2025

  • HSI-VAR: Rethinking Hyperspectral Restoration through Spatial-Spectral Visual Autoregression

    Xiangming Wang, Benteng Sun, Yungeng Liu, Haijin Zeng, Yongyong Chen, Jingyong Su, Jie Liu

    Under Review

    Key contribution: Cross-dimensional autoregressive modeling for spatiotemporal-consistent hyperspectral restoration.

  • NTIRE 2025 the 2nd Restore Any Image Model (RAIM) in the Wild Challenge [Third Prize]

    Jie Liang, Radu Timofte, et al. (Xiangming Wang)

    CVPR Workshops, 2025

  • MIPI 2025 Challenge on Deblurring for Hybrid EVS Camera: Methods and Results

    Yaqi Wu, Zhihao Fan, Hirotaka Shinozaki, et al. (Xiangming Wang)

    ICCV Workshops, 2025

Awards

Honors
  • National First Prize & Champion of the Challenge Award, National College Student IoT Design Competition.
  • Sheng Li Scholarship, Harbin Institute of Technology, top 10 students of the university, 10000 RMB.
  • Tat-Seng Chua Scholarship, Harbin Institute of Technology, top 14 students of the university, 25000 RMB.
  • Academic Scholarship, Harbin Institute of Technology, top 5% students of the university, 22000 RMB.

Contact & Collaboration

Open to Collaboration

I am open to collaboration on multimodal models, vision-language restoration, and practical AI deployment. Reach me at xmwang28@gmail.com.

Google Scholar GitHub Resume PDF