Wei Tian
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
Wei Tian has a total of 404 co-authors. Here are the top co-authors by number of shared publications:
- Long Chen
State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China
13 shared publications - Lu Xiong
Tongji University
11 shared publications - Yuyao Huang
School of Automotive Studies, Tongji University, Shanghai, China
10 shared publications Martin Lauer
Karlsruhe Institute of Technology, Germany
7 shared publications- Martin Lauer
Karlsruhe Institute of Technology (KIT)
6 shared publications - Dongpu Cao
School of Vehicle and Mobility, Tsinghua University, Beijing, China
5 shared publications
Latest Publications
- 31 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
AMTFusion: boosting 3D object detection by adaptive multi-modal temporal fusion and augmentation
Accurate perception of surrounding 3D objects is indispensable for ensuring the safety of autonomous driving systems. Furthermore, the integration of multi-modal sensors has emerged as a pivotal research area within detection paradigms based on Bird’s Eye View. In this paper, we introduce a sparse query-based 3D object detection framework by adaptive multi-modal temporal fusion of camera and LiDAR sensor data. Initially, we leverage the spatial-aware cross-modal feature aggregation to adaptively fuse instance-level point cloud and image features. We further elaborate a sequential temporal modeling approach, which facilitates an enhanced multi-modal 3D object detection within constant spatial and temporal complexity. To mitigate the challenge posed by limited high-quality training data, we incorporate a fragment-wise multi-modal temporal data augmentation strategy, aimed at bolstering the model’s performance for detecting long-tail distributed objects. Our approach achieves an NDS detection score of 75.5% on the nuScenes test set (without test time augmentation and model ensembling). These results underscore the efficacy of our approach and highlight its superiority compared to prior state-of-the-art methods. We further show that our approach remarkably improves the accuracy of multi-object tracking on nuScenes dataset.
- 31 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
AMTFusion: boosting 3D object detection by adaptive multi-modal temporal fusion and augmentation
Accurate perception of surrounding 3D objects is indispensable for ensuring the safety of autonomous driving systems. Furthermore, the integration of multi-modal sensors has emerged as a pivotal research area within detection paradigms based on Bird’s Eye View. In this paper, we introduce a sparse query-based 3D object detection framework by adaptive multi-modal temporal fusion of camera and LiDAR sensor data. Initially, we leverage the spatial-aware cross-modal feature aggregation to adaptively fuse instance-level point cloud and image features. We further elaborate a sequential temporal modeling approach, which facilitates an enhanced multi-modal 3D object detection within constant spatial and temporal complexity. To mitigate the challenge posed by limited high-quality training data, we incorporate a fragment-wise multi-modal temporal data augmentation strategy, aimed at bolstering the model’s performance for detecting long-tail distributed objects. Our approach achieves an NDS detection score of 75.5% on the nuScenes test set (without test time augmentation and model ensembling). These results underscore the efficacy of our approach and highlight its superiority compared to prior state-of-the-art methods. We further show that our approach remarkably improves the accuracy of multi-object tracking on nuScenes dataset.
- 18 March 2026
- Published by Elsevier in Infrared Physics & Technology
- Vol. 155
3D on-road object detection based on fusion of RGB and infrared images
- 18 March 2026
- Published by Elsevier in Infrared Physics & Technology
- Vol. 155
3D on-road object detection based on fusion of RGB and infrared images
- Open Access
- 13 March 2026
- Published by Elsevier in SSRN Electronic Journal
SafeCoDrive: Cross-Task Collaborative Prediction and Planning with Dual Reference to Improve End-to-End Autonomous Driving Safety
Motion prediction and trajectory planning are interdependent tasks in autonomous driving, yet existing end-to-end systems model them either sequentially or in parallel, resulting in insufficient interaction modeling and incoherent decision-making in dynamic traffic scenarios. Furthermore, most methods rely on heuristic distance-based safety constraints that ignore the kinematic and temporal nature of collision risk. To address these limitations, we propose SafeCoDrive, a collaborative end-to-end driving framework that explicitly captures prediction-planning interaction through three key innovations. First, we introduce a Cross-Task Co-Refinement (CTCR) module that enables bidirectional information flow between prediction and planning, allowing them to iteratively constrain and inform each other within a unified inference process. Second, we propose Dual-Reference Prediction (DRP), which jointly supervises agent motion in both static global coordinates and dynamic ego-centric frames, ensuring trajectory consistency under interactive maneuvers. Third, we incorporate Kinematics-Aware Safety Optimization (KSO) that replaces geometric penalties with Time-to-Collision constraints and vectorized boundary reasoning, providing physically grounded safety supervision. Comprehensive evaluations on nuScenes demonstrate that SafeCoDrive achieves state-of-the-art planning performance with an average L2 error of 0.56m while attaining the lowest collision rates (ACR: 0.08%, CCR: 0.68%). On the Bench2Drive closed-loop benchmark, SafeCoDrive improves driving score to 46.18 (+1.8% over previous best) with a sparse backbone, and to 88.25 (+1.7%) with success rate of 70.45% (+16.0%) when integrated with hierarchical planning architectures. Importantly, consistent improvements across diverse perception backbones validate that SafeCoDrive addresses paradigm-level deficiencies rather than architecture-specific optimizations, offering a general enhancement to end-to-end autonomous driving systems.
- 13 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSafeCoDrive: Cross-Task Collaborative Prediction and Planning with Dual Reference to Improve End-to-End Autonomous Driving Safety
Motion prediction and trajectory planning are interdependent tasks in autonomous driving, yet existing end-to-end systems model them either sequentially or in parallel, resulting in insufficient interaction modeling and incoherent decision-making in dynamic traffic scenarios. Furthermore, most methods rely on heuristic distance-based safety constraints that ignore the kinematic and temporal nature of collision risk. To address these limitations, we propose SafeCoDrive, a collaborative end-to-end driving framework that explicitly captures prediction-planning interaction through three key innovations. First, we introduce a Cross-Task Co-Refinement (CTCR) module that enables bidirectional information flow between prediction and planning, allowing them to iteratively constrain and inform each other within a unified inference process. Second, we propose Dual-Reference Prediction (DRP), which jointly supervises agent motion in both static global coordinates and dynamic ego-centric frames, ensuring trajectory consistency under interactive maneuvers. Third, we incorporate Kinematics-Aware Safety Optimization (KSO) that replaces geometric penalties with Time-to-Collision constraints and vectorized boundary reasoning, providing physically grounded safety supervision. Comprehensive evaluations on nuScenes demonstrate that SafeCoDrive achieves state-of-the-art planning performance with an average L2 error of 0.56m while attaining the lowest collision rates (ACR: 0.08%, CCR: 0.68%). On the Bench2Drive closed-loop benchmark, SafeCoDrive improves driving score to 46.18 (+1.8% over previous best) with a sparse backbone, and to 88.25 (+1.7%) with success rate of 70.45% (+16.0%) when integrated with hierarchical planning architectures. Importantly, consistent improvements across diverse perception backbones validate that SafeCoDrive addresses paradigm-level deficiencies rather than architecture-specific optimizations, offering a general enhancement to end-to-end autonomous driving systems.
- 8 January 2026
- Published by Springer Nature in Automotive Innovation
- Vol. 9 (2) , 326-339
A Geometry and Semantics-Based Progressive Generation Approach for Traffic Sign Adversarial Patches
- 8 January 2026
- Published by Springer Nature in Automotive Innovation
- Vol. 9 (2) , 326-339
A Geometry and Semantics-Based Progressive Generation Approach for Traffic Sign Adversarial Patches
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 12243-12254
CrossDiffAE: Cross-Modal Masked Diffusion Autoencoding for Zero-Shot Skeleton Action Recognition
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 12243-12254
CrossDiffAE: Cross-Modal Masked Diffusion Autoencoding for Zero-Shot Skeleton Action Recognition
- Open Access
- 2 October 2025
- Published by Zeal Press in International Journal of Robotics and Automation Technology
- Vol. 10, 138-148
Research on Semantic Segmentation of Fish-Eye Images for Autonomous Driving
Fisheye cameras, valued for their wide field of view, play a crucial role in perceiving the surrounding environment of vehicles. However, there is a lack of specific research addressing the processing of significant distortion features in segmenting fish-eye images. Additionally, fish-eye images for autonomous driving face the challenge of few datasets, potentially causing over fitting and hindering the model's generalization ability. Based on the semantic segmentation task, a method for transforming normal images into fish-eye images is proposed, which expands the fish-eye image dataset. By employing the Transformer network and the Across Feature Map Attention, the segmentation performance is further improved, achieving a 55.6% mIOU on Woodscape. Additionally, leveraging the concept of knowledge distillation, the network ensures a strong generalization based on dual-domain learning without compromising performance on Woodscape (54% mIOU).
- 2 October 2025
- Published by Zeal Press in International Journal of Robotics and Automation Technology
- Vol. 10, 138-148
Open AccessResearch on Semantic Segmentation of Fish-Eye Images for Autonomous Driving
Fisheye cameras, valued for their wide field of view, play a crucial role in perceiving the surrounding environment of vehicles. However, there is a lack of specific research addressing the processing of significant distortion features in segmenting fish-eye images. Additionally, fish-eye images for autonomous driving face the challenge of few datasets, potentially causing over fitting and hindering the model's generalization ability. Based on the semantic segmentation task, a method for transforming normal images into fish-eye images is proposed, which expands the fish-eye image dataset. By employing the Transformer network and the Across Feature Map Attention, the segmentation performance is further improved, achieving a 55.6% mIOU on Woodscape. Additionally, leveraging the concept of knowledge distillation, the network ensures a strong generalization based on dual-domain learning without compromising performance on Woodscape (54% mIOU).
- Open Access
- 1 October 2025
- Published by Springer Nature
- p. 1447-1462
Defending Against Adversarial Examples with Adaptive Quantization Table-Guided Image Compression
- 1 October 2025
- Published by Springer Nature
- p. 1447-1462
Open AccessDefending Against Adversarial Examples with Adaptive Quantization Table-Guided Image Compression