Zhipeng Zhang 张志鹏

Dr. Zhipeng Zhang is currently a Tenure-Track Assistant Professor in the School of Artificial Intelligence at Shanghai Jiao Tong University since April 2025. Prior to joining SJTU, he served as a Senior Researcher at KargBot from July 2022 to March 2025, leading key AI projects in autonomous driving.

He earned his Ph.D. from the National Laboratory of Pattern Recognition (NLPR) at the Chinese Academy of Sciences (CASIA) in 2022 under the supervision of Prof. Weiming Hu. During his doctoral studies, he completed a research internship at MSRA working closely with Dr. Houwen Peng, focusing on advanced computer vision and deep learning. Upon graduation, he was awarded the "Huawei Genius Young Talent".

His recent research interests include: (1) Multimodal Perception & Video Understanding; (2) Autonomous Driving (BEV, E2E); (3) Vision-Language-Action (VLA) in Embodied AI; (4) 3D Scene Reconstruction; (5) Generative AI; (6) Model Quantization for Edge Devices.

Email  /  Google Scholar

Hiring

We are recruiting undergraduate or graduate research assistants for students applying to Ph.D./master programs starting in Fall 2026. (2026年保研直博/硕士欢迎联系，另长期招聘科研实习生，可远程或来上交) If you are passionate about AI and eager to contribute to cutting-edge research, don’t hesitate to contact me (see email above). Although I just joined SJTU this year, one of my Ph.D. whom I mentored for four years recently also obtained Huawei’s Genius Young Talent Program (华为天才少年). I am committed to supporting exceptional individuals like you and helping you achieve your academic and professional goals.

News

• [2025.02] 🎉 3 papers are accepted by CVPR2025 (2 corresponding/1 co-first author).
• [2025.02] 🎉 1 paper is accepted by ICRA2025 (first author).
• [2024.09] 🎉 1 paper is accepted by NeurIPS2024.
• [2024.07] 🎉 1 paper is accepted by ECCV2024.
• [2024.07] 🎉 1 paper is accepted by IJCV (corresponding author).
• [2024.06] 🎉 1 paper is accepted by T-PAMI (corresponding author).
• [2024.02] 🎉 1 paper is accepted by CVPR2024 (corresponding author).
• [2023.09] 🎉 1 paper is accepted as Oral by BMVC.
• [2023.02] 🎉 1 paper is accepted by CVPR2023 (co-first author).
• [2022.09] 🎉 2 papers are accepted by NeurIPS2022 (1 co-first author).
• [2022.09] 🎉 Our method won the runner-up of task 1-1 and 2nd runner-up of task 1-2 in 1st Learning and Mining with Noisy Labels Challenge.
• [2022.03] 🎉 1 paper is accepted as Oral by IJCAI2022 (co-first author).
• [2021.12] 🎉 1 paper is accepted by AAAI2022 (co-first author).
• [2021.10] 🎉 1 paper is accepted by IEEE T-IP (first author).
• [2021.09] 🎉 1 paper is accepted by IEEE T-IP (co-first author).
• [2021.07] 🎉 1 paper is accepted by ICCV2021 (first author).
• [2021.02] 🎉 1 paper is accepted by CVPR2021.
• [2020.12] 🎉 1 paper is accepted by IEEE T-IP.
• [2020.09] 🎉 Our method ranks at 2nd in short-term and real-time/1st in RGBT tracks of VOT Challenge.
• [2020.07] 🎉 1 paper is accepted by ECCV2020 (first author).
• [2019.09] 🎉 Our method ranks at 2nd in RGBT tracks of VOT Challenge.
• [2019.02] 🎉 1 paper is accepted as Oral by CVP2019 (first author).

CorrBEV improves BEV detection methods in autonomous driving by introducing vision-language multimodal prototypes.

CarGS simultaneously achieves promising performances in both scene reconstruction and novel view synthesis with a unified model, improving the quality of 3DGS.

DreamTrack shows the best performances in visual tracking by dreaming the future presentation with latent world model.

Cycer reduces false positives in BEV detection of autonomous driving by propagating results of t - 1 frame to t, which generates a mask to filter distractors in BEV representation.

A-Teachers proposes the first online asymmetric framework for semi-supervised 3D LiDAR detection.

UAD proposes the first work demonstrating that an unsupervised model can outperform supervised End-to- End autonomous driving method.

VAST is the largest visual tracking benchmark to date.

AUNet proposes the Action-Units Relation Learning framework to improve the generality of forgery (deepfake) detection.

Ocean explores an efficient anchor-free framework to improve object tracking robustness.

SiamDW is the first work to solve the performance degradation in the Siamese tracking framework when using a deeper network.

Working Experience

Assistant Professor, 2025.04

Senior Researcher, 2022.07 ~ 2025.03

Research Intern, 2019.08 ~ 2020.06

Ph.D.(直博), 2017.09 ~ 2022.07

Miscellanea