Qiuqi Ruan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Qiuqi Ruan has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Gaoyun An
Institute of Information Science, Beijing Jiaotong University, Beijing, China
2 shared publications Jiying Wu
2 shared publications- Tangwen Yang
Beijing Jiaotong University
2 shared publications - Xin Wang
Unicom Data Intelligence, China Unicom, Beijing, China and Data Science & Artificial Intelligence Research Institute, China Unicom, Beijing, China
2 shared publications - Yue Zhan
School of Computer Science and Technology, Beijing Jiaotong University, Beijing, China
2 shared publications Chao Li
School of Computer Science and Technology, National University of Defense Technology, Changsha, China
1 shared publication
Latest Publications
- 18 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
GroupSD: Self-Distillation from Intermediate ViT Layers for Generalizable Person Re-identification
Person re-identification (re-ID) methods have seen remarkable advancements in closed-world testbeds, but they fail to generalize well when deployed to unseen camera networks with domain shift. Domain generalization (DG) aims to circumvent this issue by training a model that performs well on arbitrary unseen target domains without re-training. While recent advances in Knowledge Distillation (KD) show promising performance in DG, most previous methods focus solely on the final output layers without taking into account the rich information in intermediate layers, which hinders the model from capturing nuanced and robust representations. This paper reveals the importance of intermediate layers and presents GroupSD, a lightweight Self-Distillation strategy that excavates the abundant knowledge in intermediate Vision Transformer (ViTs) layers with a simple yet effective grouping mechanism by organizing ViT layers into ensembles with similar properties. Moreover, a sub-network termed Adaptive Feature Weighting (AFW) is introduced to dynamically fuse features within each group with input-dependent weights. GroupSD requires no modification to backbone architectures and can be regarded as a simple baseline for exploring intermediate features that also enable cross-block information communication. Experiments on various DG re-ID benchmarks show that this straightforward framework achieves state-of-the-art performance while incurring marginal training cost.
- 18 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
GroupSD: Self-Distillation from Intermediate ViT Layers for Generalizable Person Re-identification
Person re-identification (re-ID) methods have seen remarkable advancements in closed-world testbeds, but they fail to generalize well when deployed to unseen camera networks with domain shift. Domain generalization (DG) aims to circumvent this issue by training a model that performs well on arbitrary unseen target domains without re-training. While recent advances in Knowledge Distillation (KD) show promising performance in DG, most previous methods focus solely on the final output layers without taking into account the rich information in intermediate layers, which hinders the model from capturing nuanced and robust representations. This paper reveals the importance of intermediate layers and presents GroupSD, a lightweight Self-Distillation strategy that excavates the abundant knowledge in intermediate Vision Transformer (ViTs) layers with a simple yet effective grouping mechanism by organizing ViT layers into ensembles with similar properties. Moreover, a sub-network termed Adaptive Feature Weighting (AFW) is introduced to dynamically fuse features within each group with input-dependent weights. GroupSD requires no modification to backbone architectures and can be regarded as a simple baseline for exploring intermediate features that also enable cross-block information communication. Experiments on various DG re-ID benchmarks show that this straightforward framework achieves state-of-the-art performance while incurring marginal training cost.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 1-16
E2M-Fusion: An End-to-end Multi-level Model for Multimodal Image Fusion with Missing Data
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 1-16
E2M-Fusion: An End-to-end Multi-level Model for Multimodal Image Fusion with Missing Data
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 1-15
CUDA-CC: Continual Unsupervised Domain Adaptation for Crowd Counting
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 1-15
CUDA-CC: Continual Unsupervised Domain Adaptation for Crowd Counting
- 27 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 9749-9762
TG-Pose: Delving Into Topology and Geometry for Category-Level Object Pose Estimation
Category-level 6D object pose estimation aims to estimate the pose and size of unseen objects with known categories. Existing methods mainly focus on capturing geometric features to handle shape variations, and are prone to failure in occlusion and noisy environments. In this paper, we propose TG-Pose, a unified pose estimation framework that delves into topology and geometry to deal with the above issues. To exploit topological properties, we first propose a topological feature predictor and a topological label generator to dig into the underlying structural details from encoded features using persistent homology. Then, the topological and geometric features are employed to facilitate the symmetry reconstruction of the original point cloud to obtain a reliable and coherent object shape, which, in turn, guides the pose estimation. For each object category, we construct geometric and topological templates by leveraging inherent intra-class similarities. These templates enhance the reliability of pose estimation and the completeness of object structure through geometric alignment and topological guidance, especially when handling incomplete objects. Moreover, a pose-aware enhancement strategy is designed to enhance the encoder in learning pose-sensitive features and robustness to noisy point clouds. Experimental results show that TG-Pose outperforms the state-of-the-art solutions on public benchmarks and achieves better generalization in real-world datasets. Project Page https://sites.google.com/view/tg-pose.
- 27 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 9749-9762
TG-Pose: Delving Into Topology and Geometry for Category-Level Object Pose Estimation
Category-level 6D object pose estimation aims to estimate the pose and size of unseen objects with known categories. Existing methods mainly focus on capturing geometric features to handle shape variations, and are prone to failure in occlusion and noisy environments. In this paper, we propose TG-Pose, a unified pose estimation framework that delves into topology and geometry to deal with the above issues. To exploit topological properties, we first propose a topological feature predictor and a topological label generator to dig into the underlying structural details from encoded features using persistent homology. Then, the topological and geometric features are employed to facilitate the symmetry reconstruction of the original point cloud to obtain a reliable and coherent object shape, which, in turn, guides the pose estimation. For each object category, we construct geometric and topological templates by leveraging inherent intra-class similarities. These templates enhance the reliability of pose estimation and the completeness of object structure through geometric alignment and topological guidance, especially when handling incomplete objects. Moreover, a pose-aware enhancement strategy is designed to enhance the encoder in learning pose-sensitive features and robustness to noisy point clouds. Experimental results show that TG-Pose outperforms the state-of-the-art solutions on public benchmarks and achieves better generalization in real-world datasets. Project Page https://sites.google.com/view/tg-pose.
- 25 February 2014
- Published by Elsevier in Neurocomputing
- Vol. 138, 283-296
Dimensionality reduction using graph-embedded probability-based semi-supervised discriminant analysis
- 25 February 2014
- Published by Elsevier in Neurocomputing
- Vol. 138, 283-296
Dimensionality reduction using graph-embedded probability-based semi-supervised discriminant analysis
- 25 August 2012
- Published by Elsevier in Neurocomputing
- Vol. 99, 592-602
Enhancing sparsity via ℓp (0<p<1) minimization for robust face recognition
- 25 August 2012
- Published by Elsevier in Neurocomputing
- Vol. 99, 592-602
Enhancing sparsity via ℓp (0<p<1) minimization for robust face recognition
- 1 April 2012
- Published by Elsevier in Neurocomputing
- Vol. 82, 99-108
A remarkable standard for estimating the performance of 3D facial expression features
- 1 April 2012
- Published by Elsevier in Neurocomputing
- Vol. 82, 99-108
A remarkable standard for estimating the performance of 3D facial expression features
- 1 July 2011
- Published by Elsevier in Pattern Recognition
- Vol. 44 (7) , 1497-1513
Orthogonal Tensor Neighborhood Preserving Embedding for facial expression recognition
- 1 July 2011
- Published by Elsevier in Pattern Recognition
- Vol. 44 (7) , 1497-1513
Orthogonal Tensor Neighborhood Preserving Embedding for facial expression recognition
- 1 January 2010
- Published by Vilnius University Press in Informatica
- Vol. 21 (2) , 259-276
Exemplar-Based Image Completion Model Employing PDE Corrections
Publisher: Vilnius University Institute of Data Science and Digital Technologies, Journal: Informatica, Title: Exemplar-Based Image Completion Model Employing PDE Corrections, Authors: Jiying Wu, Qiuqi Ruan, Gaoyun An , To suitably complete an image without seams, block effects, and artifacts, a novel exemplar-based image completion model employing partial differential equation (PDE) is proposed. Firstly, the novel model determines the processing order of exemplar according to a composite function, which is the product of the colour property and structure property in exemplar. Then the exemplar along a geometric structure is processed prior to other parts of the image. Secondly, the most similar exemplar is found in the CIELAB colour space, and the size of it is adaptively determined by the local textured information. Thirdly, a Poisson equation is adopted to remove the seams, block effects, and artifacts in the image generated by the exemplar-based model. Finally, a bi-directional diffusion PDE is used to assist the completion of lathy linear structure. Experimental results demonstrate that the novel model can properly reconstruct the target region while preserving the geometric structure without inducing block effects, which leads to its better performance than the conventional exemplar-based image completion models.
- 1 January 2010
- Published by Vilnius University Press in Informatica
- Vol. 21 (2) , 259-276
Exemplar-Based Image Completion Model Employing PDE Corrections
Publisher: Vilnius University Institute of Data Science and Digital Technologies, Journal: Informatica, Title: Exemplar-Based Image Completion Model Employing PDE Corrections, Authors: Jiying Wu, Qiuqi Ruan, Gaoyun An , To suitably complete an image without seams, block effects, and artifacts, a novel exemplar-based image completion model employing partial differential equation (PDE) is proposed. Firstly, the novel model determines the processing order of exemplar according to a composite function, which is the product of the colour property and structure property in exemplar. Then the exemplar along a geometric structure is processed prior to other parts of the image. Secondly, the most similar exemplar is found in the CIELAB colour space, and the size of it is adaptively determined by the local textured information. Thirdly, a Poisson equation is adopted to remove the seams, block effects, and artifacts in the image generated by the exemplar-based model. Finally, a bi-directional diffusion PDE is used to assist the completion of lathy linear structure. Experimental results demonstrate that the novel model can properly reconstruct the target region while preserving the geometric structure without inducing block effects, which leads to its better performance than the conventional exemplar-based image completion models.
- 1 January 2009
- Published by Vilnius University Press in Informatica
- Vol. 20 (1) , 151-163
A Joint-Diffused Inpainting Model for Underexposure Image Preserving the Linear Geometric Structure
Publisher: Vilnius University Institute of Data Science and Digital Technologies, Journal: Informatica, Title: A Joint-Diffused Inpainting Model for Underexposure Image Preserving the Linear Geometric Structure, Authors: Jiying Wu, Qiuqi Ruan, Gaoyun An , To restore the underexposure image, an illumination compensation inpainting model which employs the joint-diffused partial differential equations (PDEs) is proposed. Firstly, the novel model compensates the illumination effect in multi-scaled underexposure images respectively. Secondly, the information in the fused compensated image is restored by PDEs which diffuse the geometric property and gray information into the target region simultaneously. Experimental results demonstrate that the novel model can properly restore scratches while compensating the illumination effect in underexposure image, and the joint-diffused PDEs which are employed in it lead to a better performance than the conventional PDE inpainting models.
- 1 January 2009
- Published by Vilnius University Press in Informatica
- Vol. 20 (1) , 151-163
A Joint-Diffused Inpainting Model for Underexposure Image Preserving the Linear Geometric Structure
Publisher: Vilnius University Institute of Data Science and Digital Technologies, Journal: Informatica, Title: A Joint-Diffused Inpainting Model for Underexposure Image Preserving the Linear Geometric Structure, Authors: Jiying Wu, Qiuqi Ruan, Gaoyun An , To restore the underexposure image, an illumination compensation inpainting model which employs the joint-diffused partial differential equations (PDEs) is proposed. Firstly, the novel model compensates the illumination effect in multi-scaled underexposure images respectively. Secondly, the information in the fused compensated image is restored by PDEs which diffuse the geometric property and gray information into the target region simultaneously. Experimental results demonstrate that the novel model can properly restore scratches while compensating the illumination effect in underexposure image, and the joint-diffused PDEs which are employed in it lead to a better performance than the conventional PDE inpainting models.