Ruibin Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ruibin Wang has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Bowei Xing
State Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University, Beijing, China
5 shared publications - Xianghua Ying
State Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University, Beijing, China
5 shared publications - Ruohao Guo
State Key Laboratory of General AI Peking University 100871 Beijing
2 shared publications Bowei Xing
1 shared publication- Ji Shi
State Key Laboratory of General AI Peking University 100871 Beijing
1 shared publication - Jian Jun Zhang
Bournemouth University, Bournemouth, United Kingdom
1 shared publication
Latest Publications
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-15
Improving Out-of-Distribution Generalization on Point Clouds with Cross-Domain Adversarial Distillation
Point cloud is an important data format to represent 3D objects and real world scenes, which is widely used in many fields such as virtual reality and autonomous driving. In recent years, various methods have been proposed for point cloud recognition and achieved great performance. However, since these methods are always under the i.i.d. assumption, the trained model inevitably suffers drastic performance degradation when evaluated on out-of-distribution (OOD) data. To address the above problem, in this paper, we propose a framework named cross-domain adversarial distillation, with the goal of learning a model that generalizes well on OOD point clouds. Concretely, we initiate by dynamically generating adversarial point cloud examples that deviate from original domain and exhibit diversification, so as to progressively broaden the training distribution. Subsequently, we implement a bidirectional distillation scheme to ensure semantic consistency between the generated novel-domain example and source domain example, which facilitates the exchange of cross-domain knowledge through teacher-student cooperation. By iteratively alternating between the above processes, the model is encouraged to learn more generalizable representations when encountering point clouds from novel distributions, leading to improved performance on the unseen target domain. We carry experiments on two typical datasets, ModelNet40-C for corrupted point clouds and Sim-to-Real for real-world point clouds, implemented with different 3D recognition models. The experimental results show that our method achieves better performance on these two benchmarks, which demonstrate the effectiveness of our method for improving point cloud OOD generalization.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-15
Improving Out-of-Distribution Generalization on Point Clouds with Cross-Domain Adversarial Distillation
Point cloud is an important data format to represent 3D objects and real world scenes, which is widely used in many fields such as virtual reality and autonomous driving. In recent years, various methods have been proposed for point cloud recognition and achieved great performance. However, since these methods are always under the i.i.d. assumption, the trained model inevitably suffers drastic performance degradation when evaluated on out-of-distribution (OOD) data. To address the above problem, in this paper, we propose a framework named cross-domain adversarial distillation, with the goal of learning a model that generalizes well on OOD point clouds. Concretely, we initiate by dynamically generating adversarial point cloud examples that deviate from original domain and exhibit diversification, so as to progressively broaden the training distribution. Subsequently, we implement a bidirectional distillation scheme to ensure semantic consistency between the generated novel-domain example and source domain example, which facilitates the exchange of cross-domain knowledge through teacher-student cooperation. By iteratively alternating between the above processes, the model is encouraged to learn more generalizable representations when encountering point clouds from novel distributions, leading to improved performance on the unseen target domain. We carry experiments on two typical datasets, ModelNet40-C for corrupted point clouds and Sim-to-Real for real-world point clouds, implemented with different 3D recognition models. The experimental results show that our method achieves better performance on these two benchmarks, which demonstrate the effectiveness of our method for improving point cloud OOD generalization.
- 19 June 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 293
Improving Bird’s Eye View based 3D object detection via Learnable Perspective View
- 19 June 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 293
Improving Bird’s Eye View based 3D object detection via Learnable Perspective View
- 1 November 2024
- Published by Springer Nature
- p. 393-409
Hierarchical Unsupervised Relation Distillation for Source Free Domain Adaptation
- 1 November 2024
- Published by Springer Nature
- p. 393-409
Hierarchical Unsupervised Relation Distillation for Source Free Domain Adaptation
- 11 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 4527-4539
Exploiting Temporal Correlations for 3D Human Pose Estimation
Exploiting the rich temporal information in human pose sequences to facilitate 3D pose estimation has garnered particular attention. While various learning architectures have been designed for temporal exploiting, these architectures are usually trained via the 3D pose loss independently imposed on every single frame, without explicit temporal signals introduced for supervision. This inevitably increases the difficulty of temporal exploiting, since the network must reason about the meaningful temporal information based on the non-temporal single-frame supervision first. Only then, the network can utilize this information to guide sequence modeling. Recently, some work introduce temporal smoothness as an explicit supervision signal, which makes the network more straightforwardly reaches the temporal information from the supervision signal, thus improving the temporal exploiting. However, the temporal smoothness only roughly measures the short-term temporal properties between adjacent frame pairs. In this work, we propose to generalize the supervision of temporal smoothness to temporal correlations, letting the network precisely consider more comprehensive temporal properties in sequences. We contribute two novel correlation-based loss functions, which adopt different strategies to respectively regularize the encoder and decoder sides of the network for temporal exploiting. Besides, we design a pre-training scheme to ensure a general convergence of existing pose estimators under our correlation losses. Experiments on three benchmarks demonstrate that our method can be compatible with different networks, improving their temporal exploiting ability to output more accurate and robust pose estimations.
- 11 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 4527-4539
Exploiting Temporal Correlations for 3D Human Pose Estimation
Exploiting the rich temporal information in human pose sequences to facilitate 3D pose estimation has garnered particular attention. While various learning architectures have been designed for temporal exploiting, these architectures are usually trained via the 3D pose loss independently imposed on every single frame, without explicit temporal signals introduced for supervision. This inevitably increases the difficulty of temporal exploiting, since the network must reason about the meaningful temporal information based on the non-temporal single-frame supervision first. Only then, the network can utilize this information to guide sequence modeling. Recently, some work introduce temporal smoothness as an explicit supervision signal, which makes the network more straightforwardly reaches the temporal information from the supervision signal, thus improving the temporal exploiting. However, the temporal smoothness only roughly measures the short-term temporal properties between adjacent frame pairs. In this work, we propose to generalize the supervision of temporal smoothness to temporal correlations, letting the network precisely consider more comprehensive temporal properties in sequences. We contribute two novel correlation-based loss functions, which adopt different strategies to respectively regularize the encoder and decoder sides of the network for temporal exploiting. Besides, we design a pre-training scheme to ensure a general convergence of existing pose estimators under our correlation losses. Experiments on three benchmarks demonstrate that our method can be compatible with different networks, improving their temporal exploiting ability to output more accurate and robust pose estimations.
- 28 June 2023
- Published by Elsevier in Pattern Recognition
- Vol. 144
Improving point cloud classification and segmentation via parametric veronese mapping
- 28 June 2023
- Published by Elsevier in Pattern Recognition
- Vol. 144
Improving point cloud classification and segmentation via parametric veronese mapping
- Open Access
- 7 March 2019
- Published by Springer Nature in Optimization and Engineering
- Vol. 20 (3) , 907-928
Optimal NURBS conversion of PDE surface-represented high-speed train heads
- 7 March 2019
- Published by Springer Nature in Optimization and Engineering
- Vol. 20 (3) , 907-928
Open AccessOptimal NURBS conversion of PDE surface-represented high-speed train heads