Song Wang

I am a Ph.D. student in the College of Computer Science and Technology at Zhejiang University, advised by Prof. Jianke Zhu . Prior to that, I obtained my B.Eng. at Zhejiang University, supervised by Prof. Zhiwei Xu and Prof. Hangfang Zhao .

I am interested in computer vision and machine learning. My current research focuses on:

  • Vision-language models for visual tasks and applications.
  • Vision-centric occupancy network a.k.a semantic scene completion.
  • Label-efficient learning for 2D and 3D dense prediction.
  • Various forms of academic collaboration and discussion are welcome. Feel free to reach out!

    Email  /  Google Scholar  /  Github

    profile photo

    Publications

    * indicates equal contribution

    dise OccFiner: Offboard Occupancy Refinement with Hybrid Propagation
    Hao Shi*, Song Wang*, Jiaming Zhang, Xiaoting Yin, Zhongdao Wang, Zhijian Zhao, Guangming Wang, Jianke Zhu, Kailun Yang, Kaiwei Wang
    arXiv, 2024
    [arXiv] [Code]

    In addressing the challenges of inferior performance and data closure in vision-based SSC, we introduce OccFiner, the first offboard SSC setup to solve the unreliability of the onboard model.

    dise Label-efficient Semantic Scene Completion With Scribble Annotations
    Song Wang, Jiawei Yu, Wentong Li, Hao Shi, Kailun Yang, Junbo Chen, Jianke Zhu
    International Joint Conference on Artificial Intelligence (IJCAI), 2024
    [arXiv] [Code]

    In this work, we have presented a scribble-based label-efficient benchmark ScribbleSC for semantic scene completion in autonomous driving. To enhance the performance in this setting, an effective scribble-supervised approach Scribble2Scene has been developed.

    dise Not All Voxels Are Equal: Hardness-Aware Semantic Scene Completion with Self-Distillation
    Song Wang, Jiawei Yu, Wentong Li, Wenyu Liu, Xiaolu Liu, Junbo Chen, Jianke Zhu
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024
    [arXiv] [Code]

    In this paper, we adhere to the principle of not all voxels are equal and propose hardness-aware semantic scene completion (HASSC).

    dise MGMap: Mask-Guided Learning for Online Vectorized HD Map Construction
    Xiaolu Liu, Song Wang, Wentong Li, Ruizi Yang, Junbo Chen, Jianke Zhu
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024
    [arXiv] [Code]

    In this paper, we propose MGMap, an effective approach to online HD map vectorization with the guidance of learned masks.

    dise Label-efficient Segmentation via Affinity Propagation
    Wentong Li*, Yuqian Yuan*, Song Wang, Wenyu Liu, Dongqi Tang, Jian Liu, Jianke Zhu, Lei Zhang
    Conference on Neural Information Processing Systems (NeurIPS), 2023
    [arXiv] [Code] [Project Page]

    We propose a method named APro, designed to generate precise soft pseudo labels online for unlabeled regions within segmentation networks.

    dise Point2Mask: Point-supervised Panoptic Segmentation via Optimal Transport
    Wentong Li, Yuqian Yuan, Song Wang, Jianke Zhu, Jianshu Li, Jian Liu, Lei Zhang
    IEEE/CVF International Conference on Computer Vision (ICCV), 2023
    [arXiv] [Code]

    In this paper, we present an effective method, namely Point2Mask, to achieve highquality panoptic prediction using only a single random point annotation per target for training.

    dise LiDAR2Map: In Defense of LiDAR-Based Semantic Map Construction Using Online Camera Distillation
    Song Wang, Wentong Li, Wenyu Liu, Xiaolu Liu, Jianke Zhu
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023
    [arXiv] [Code]

    In this work, an efficient semantic map construction framework named LiDAR2Map, is presented with an effective BEV feature pyramid decoder and an online Camera-to-LiDAR distillation scheme.

    dise Meta-RangeSeg: LiDAR Sequence Semantic Segmentation Using Multiple Feature Aggregation
    Song Wang, Jianke Zhu, Ruixiang Zhang
    IEEE Robotics and Automation Letters (RA-L with IROS, IF: 5.2), 2022
    [arXiv] [Code]

    We propose a novel approach to LiDAR semantic segmentation, which introduces a range residual image representation to capture the spatial-temporal information.

    Honors and Awards

  • China's Optics Valley Scholarship, Donghu New Technology Development Zone
  • Graduate with Merit A Performance, Zhejiang University
  • Award of Honor for Graduate, Zhejiang University
  • Outstanding Undergraduate Award, Zhejiang University
  • Zhejiang Provincial Government Scholarship, Zhejiang Province
  • Zhongtian Technology First-Class Scholarship, ZTT Group
  • Zhejiang University Scholarship - Second Prize, Zhejiang University
  • First Prize in Advanced Mathematics Competition, Zhejiang Province
  • Academic Services

  • Conference Reviewer: NeurIPS 2024, ECCV 2024, SynData4CV@CVPR 2024, ACM MM 2023-2024, ICRA 2024
  • Journal Reviewer: IEEE Robotics and Automation Letters, IEEE Transactions on Intelligent Vehicles

  • Website Template


    © Song Wang | Last updated: May 24, 2024