Tianrui Liu(刘天瑞)

I am currently an Associate Professor in the Pattern Recognition and Machine Intelligence (PRMI) group at the Department of Computer Science, National University of Defense Technology (NUDT). Before, I was a Research Associate in the BioMedIA Group in Imperial College London, advised by Professor Daniel Rueckert and Dr. Bernhard Kainz.

I'm interested in computer vision, deep learning, object detection and pattern recognition, image and video processing and medical image analysis.

In my spare time, I like playing piano, reading and playing badminton.

Email  /  CV  /  Google Scholar  /  LinkedIn

• Ph.D. in Computer Vision and Image Processing, Imperial College London
  Nov. 2015 - Nov. 2019
• M.Phil. in Image and Video Processing, University of Hong Kong
  Sep. 2013 - Nov. 2015
• B.Eng. (First Class Honours) in EIE, Hong Kong Polytechnic University
  Sep. 2011 - Jun. 2013
• B.Eng. in Microelectronics, San Yat-Sen University
  Sep. 2009 - Jun. 2011

• Resesarch Associate @ Imperial College London.
  Aug. 2019 - Aug. 2021
• Research Intern @ Tencent AI Lab (Shenzhen, China), advised by Dr. Wenhan Luo
  Oct. 2019 - Apr. 2020
• Visiting Research Scholar @ Hong Kong Polytechnic University, advised by Prof. Kenneth K. M. Lam
  Aug. 2018
• Research Intern @ RISA Sicherheitsanalysen GmbH (Berlin, Germany)
  June. 2018 - Sep. 2018
• Visiting Research Scholar @ Hong Kong Polytechnic University, advised by Prof. Wan-Chi Siu
  Jun. 2016 - Aug. 2016

[2025.07] Our work "VSumMamba: Mamba Empowered Efficient Video Summarization with Multi-Scale Spatial-Temporal Modeling" has been accepted by ACM Multimedia (ACM MM) 2025   (CCF A).

[2025.06] Our work "Sampling Enhanced Contrastive Multi-View Remote Sensing Data Clustering With Long-Short Range Information Mining" has been accepted by IEEE Transactions on Knowledge and Data Engineering (IEEE TKDE)   (CCF A).

[2025.06] Our work "Generalized Deep Multi-view Clustering via Causal Learning with Partially Aligned Cross-view Correspondence" has been accepted by International Conference on Computer Vision 2025 (ICCV)   (CCF A).

[2025.03] Our work "EASEMVC: Efficient Dual Selection Mechanism for Deep Multi-View Clustering" has been accepted by IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2025   (CCF A).

[2025.03] Our work "A Lightweight Deep Exclusion Unfolding Network for Single Image Reflection Removal" has been accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence(IEEE TPAMI)   (CCF A).

[2025.03] Our work "SMILENet: Unleashing Extra-Large Capacity Image Steganography via a Synergistic Mosaic Invertible Hiding Network" is now available on arXiv.

[2025.03] Four of our papers have been accepted by IEEE International Conference on Multimedia and Expo (ICME) 2025.

The paper introduces a lightweight deep exclusion unfolding network that effectively removes reflections from single images via a resource-efficient unfolding architecture integrating exclusion mechanisms, achieving high-quality restoration with reduced computational complexity.

This paper proposes SEC-LSRM, a spatial- and sampling-enhanced contrastive framework that unifies long/short-range information mining and idempotent-guided sampling to set new state-of-the-art for remote-sensing multi-view clustering.

This paper proposes SMILENet, a synergistic invertible-CNN framework that, for the first time, enables distortion-free hiding and recovery of up to 25 images while introducing a capacity-distortion metric to set the new state-of-the-art in ultra-high-capacity image steganography.

We propose SGROD, the first SAM-guided open-world detector that boosts unknown-object recall by ≈20 % while preserving known-object precision via dynamic label assignment, cross-layer learning and SAM-based negative sampling.

This paper introduces EASEMVC, a dual-selection mechanism that adaptively chooses view pairs via optimal-transport graphs and reweights samples for noise-robust contrastive learning, setting new state-of-the-art results in deep multi-view clustering.

We propose AIMC-CVR, which alleviates anchor-shift in incomplete multi-view clustering by cross-view anchor learning and affine combination reconstruction, outperforming SOTA on seven datasets.

This paper proposes AIM-VR, an all-in-one video restoration framework that unifies rain/haze/snow/noise/blur removal via dual-path Mamba with Hilbert scanning and frequency-adaptive fusion, achieving state-of-the-art quality with Vision-Mamba efficiency.

We propose VSumMamba, a Mamba-empowered video summarization framework that integrates a cascaded temporal module and a parallel spatial module with three hierarchical multi-scale schemes to achieve state-of-the-art accuracy while maintaining linear complexity and low computational cost.

This paper proposes TGM-MVC, a tree-based multi-view clustering framework that explicitly preserves view gaps via minimum spanning trees to retain cross-view diversity while enhancing consensus, achieving state-of-the-art performance on six benchmarks.

This paper presents CAM, a parameter-free Category Alignment Mechanism that lets each query sample pick category-specific features by exploiting intra-/inter-category contrasts, yielding consistent gains on six few-shot benchmarks without extra training.

We propose a multi-mini-patch adversarial attack for remote sensing image classification, generating small patches via adaptive diversity-aware optimization, boosting stealthiness and attack success with practical deployment.

We propose 3DST-UNet-RL, a deep RL-based framework using a 3D spatio-temporal U-Net for video summarization. We extend our video summarization method for ultrasound scanning videos, where automated report generation with short but relevant video clips is desirable.

In this paper, we propose a gated multi-layer convolutional feature extraction method which can adaptively generate discriminative features for candidate pedestrian regions.

We propose an ultrasound video summarization method to summarize the long examination videos. The proposed method can remove parts that are not relevant for diagnostics and meanwhile guarantees the preservation of decisive diagnostic information.

we propose a novel multi-view motion estimation network (MulViMotion), which integrates 2D cine CMR images acquired in short-axis and long-axis planes to learn a consistent 3D motion field of the heart.

We piloted the end-to-end automation of the mid-trimester screening ultrasound scan using AI enabled tools. Survey responses suggest that the AI toolshelped sonographers to concentrate on image interpretation by removing disruptivetasks.

We exploits different combination of multi-resolution CNN features for pedestrian candidates of different scales.

Our method extends the Faster R-CNN detection framework by adding a branch of network for semantic image segmentation.

A novel hierarchical random forests based super-resolution (SRHRF) method is proposed to learn statistical priors from external training images.

We propose an effective image labeling method with a hierarchical framework consists of two layers of random forests (RF). In the first layer, RF is performed on superpixel basis. In the second layer, structured RF is applied to make use of the topological distribution of the object classes.

We propose a two-stage framework for fast image interpolation via random forests. The underlying idea of this proposed work is to apply random forests to classify the natural image patch space into numerous subspaces and learn a linear regression model for each subspace to map the low-resolution image patch to high-resolution image patch.

• Chinese Government Award for Outstanding (Non-Government Sponsored) Students Abroad
  2020
• Imperial College Department Scholarship (£130,000), Imperial College London
  2016 - 2019
• Young Scientist Award, IEEE Region 10 Conference TENCON
  2015
• Dean's List of Outstanding Students, HKPolyU
  2012 & 2013
• Outstanding Performance Scholarship (HK$ 80,000), HKSAR Government
  2012
• PolyU EIE (Non-local Student) Scholarship (HK$ 100,000), HKPolyU
  2011

• Convex Optimization
• Machine Learning
• Pratical AI Computing

Journal Reviewer:

• IEEE Transactions on Image Processing (TIP)
• IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
• IEEE Transactions on Medical Imaging (TMI)
• Information Fusion
• Transactions on Geoscience and Remote Sensing (TGRS)
• IEEE Transactions on Multimedia (TMM)
• IEEE Internet of Things Journal (IoT)
• Future Generation Computer Systems
• Pattern Recognation, etc.

• Deep learning in Pedestrian Detection and Video Analysis
  @Huazhong University of Science and Technology, WuHan, China
  Dec. 2020
• Deep Learning for Pedestrian Detection and Medical Video Summarization
  @Nanjing University of Aeronautics and Astronautics, NanJing, China
  Sep. 2020
• What Can Help Pedestrian Detection?
  @National University of Defense Technology, ChangSha, China
  Dec. 2019
• Deep Learning for Pedestrian Detection under Complex Environment
  @NorthEast University，Department of Computer Science, ShenYang, China
  Dec. 2019
• Enhanced Pedestrian Detection using Deep Learning based Semantic Segmentation and Scale-Aware Scheme
  @National Technical University of Athens, Athens, Greece
  Sept. 2018
• Low Resolution Face Detection for Video Surveillance Technologies
  @Artificial Intelligence and Mulitmedia Lab, HK PolyU, HK
  Sept. 2018