Junbo Yin is a Ph.D. candidate in Beijing Lab of Intelligent Information Technology at Beijing Institute of Technology (BIT), under the supervision of Prof. Jianbing Shen. He is also a visiting Ph.D. at EPFL, supervised by Prof. Pascal Frossard. His research interests include 3D object detection, self-supervised learning, and 3D domain adaptation. He has published 15 conference and journal papers such as CVPR, ICCV, ECCV and TPAMI, with oevr 1050 citations on Google Scholar. He has obtained many prestigious scholarships and honorary titles, including Excellent Graduate Student, Special Grad Scholarship and Huawei Scholarship. Additionally, he has been awarded by the Zhejiang Lab’s International Talent Fund for Young Professionals.
Ph.D. in Computer Vision, 2018
Beijing Institute of Technology
MEng in Pattern Recognition and Intelligent System, 2016
Beijing Institute of Technology
BSc in Automation, 2012
North China Electric Power University
We present a novel learning paradigm, LWSIS, that inherits the fruits of off-the-shelf 3D point cloud to guide the training of 2D instance segmentation models to save mask-level annotations. We advocate a new dataset nuInsSeg based on nuScenes to extend existing 3D LiDAR annotations with 2D image segmentation annotations.
We prsent SSDA3D in this work, which is the first effort for semi-supervised domain adaptation in the context of 3D object detection. SSDA3D is achieved by a novel framework that jointly addresses inter-domain adaptation and intra-domain generalization.
We propose a proposal-level point cloud SSL framework, named ProposalContrast, that conducts proposal-wise contrastive pre-training for detection-aligned representation learning. We comprehensively demonstrate the generalizability of our model across diverse 3D detector architectures and datasets.
We propose ProficientTeachers for LiDAR-based 3D object detection, which is achieved by promoting the plain teacher model to proficient teachers inspired by ensemble learning. Our framework not only performs better results, but also removes the necessity of confidencebased thresholds for filtering pseudo labels.
A general point cloud-based 3D video object detection framework is proposed by leveraging both short-term and long-term point cloud information. We build the proposed 3D video object detectio framework upon both anchor-based and anchor-fre 3D object detectors. Also, the proposed framework can be easily deployed in both online and offlin modes.
Reviewer: