Photo

I am currently a senior undergraduate student majoring in Intelligent Science and Technology (Excellence Class) at Fudan University . I received the Bachelor of Engineering degree in June 2023. I am now a first-year Ph.D. student in Aggie Graphics Group, Texas A&M University, and my advisors are Prof. Wenping Wang and Prof. Xin Li. I also worked as a research assistant at HKUST CSE department with Prof. Dan Xu from Feb. 2022 to June 2023. Previously, I have been working with Prof. Tao Chen and Prof. Jiayuan Fan at Fudan University.

My research interests lie in computer vision, computer graphics and multi-task scene understanding. For computer vision, I work on representation learning and knowledge transferring between various tasks, in order to build robust and versatile multi-task models. For computer graphics, I work on 3D model reconstruction and parameterization, including parametric surface / volume, neural 3D representations, etc.

Email: jdzhang [at] tamu [dot] edu / jdzhang19 [at] fudan [dot] edu [dot] cn
Wechat: Triumph0929
OpenReview / GitHub / Google Scholar / CV


News



Education


Texas A&M University
Department of Computer Science and Engineering
Ph.D Student
Auguest 2023 - recent

Fudan University
Intelligent Science and Technology (excellent class), School of Information Science and Engineering
Undergraduate Student
September 2019 - June 2023, received the bachelor's degree in Jun 2023



Internship


Tencent America
Research Intern
May 2024 - Aug 2024



Publications


Jingdong Zhang*, Hanrong Ye, Wenping Wang, Xin Li, Dan Xu
Learning Hierarchical Task Tokens for Effective Multi-Task Partially Annotated Dense Predictions
CVPR 2024 (IEEE Conference on Computer Vision and Pattern Recognition), Under Review
Abstract: This research proposes a new approach to multi-task dense predictions with partially labeled data. We introduction hierarchical task tokens (HiTTs) to capture multi-level representations. The global task tokens conduct cross-task interactions and transfer knowledge from labeled to unlabeled tasks, while the fine-grained task tokens form high-quality task predictions at a finer granularity.

Jingdong Zhang, Jiayuan Fan*, Peng Ye, Bo Zhang, Hancheng Ye, Baopu Li, Yancheng Cai, Tao Chen
Rethinking of Feature Interaction for Multi-task Learning on Dense Prediction
TPAMI (IEEE Transactions on Pattern Analysis and Machine Intelligence), Under Revision    [ arxiv ]
Abstract: This work introduces a new Bridge-Feature-Centric Interaction (BFCI) for multi-task learning on dense predictions. It uses a Bridge Feature Extractor (BFE) to create strong bridge features and a Task Pattern Propagation (TPP) to solve the task-pattern entanglement issue, which results in task-specific features with higher quality and discriminative task representations. A Task-Feature Refiner (TFR) is then used to refine the final task predictions.

Yancheng Cai, Bo Zhang, Baopu Li, Tao Chen*, Hongliang Yan, Jingdong Zhang, Jiahao Xu
Rethinking Cross-Domain Pedestrian Detection: A Background-Focused Distribution Alignment Framework for Instance-Free One-Stage Detectors
TIP (IEEE Transactions on Image Processing)    [ paper ]    [ code ]
Abstract: We introduce a new approach for cross-domain pedestrian one-stage detectors. The paper identifies a foreground-background misalignment issue in image-level feature alignment, and a novel framework, Background-Focused Distribution Alignment (BFDA) is proposed to address this issue.



Research Experience



Selected Award



Academic Service



Wukang Rd, Shanghai
Mount Dongda, Tibet
Bell Tower, Xi'an