profile photo

Stone Tong ZHANG

I am a full-time researcher at CUHK-SZ with Prof. Haizhou Li . I gained MS degree from University of California, Irvine and Bachelor of Computer Science and Engineering from Southern University of Science and Technology. My research interests include Computer Vision and Multimodal Learning. Feel free to contact me at any time.

Email Github Linkedin CV

  • Sep. 2019-June 2023, "Southern University of Science and Technology", BEng, Computer Science and Engineering

  • Sep. 2022-June 2024, "University of California, Irvine", M.S., Computer Engineering

    • Academic Projects

    Unintended Side Effects of Defense Mechanisms in Large Language Models

    Prof. Haizhou Li, CUHK-Shenzhen
    PyTorch, LaTeX
  • Conducted a comprehensive analysis of 11 defense mechanisms applied to 6 LLMs, evaluating their impact on model performance, over-refusal, and token overhead
  • Proposed 9 meta-defenders to systematically analyze trade-offs between safety and utility in model responses
  • Provided actionable insights on designing robust LLMs for safety-critical applications

    • Human-Readable SVG Generation with Vision Language Models

    Assistant Prof. Haohan Wang, UIUC
    PyTorch
  • Developed S²VG², an innovative method integrating vision language models for scalable vector graphics (SVG) generation
  • Curated the SVG-SHAPE dataset to benchmark SVG generation and model reasoning capabilities
  • Demonstrated state-of-the-art performance in SVG reasoning of LLMs and vision metrics

    • One-shot Controllable Head Avatar Creation

    Prof. Xiaohui Xie, UCI
    PyTorch
  • Pioneered CVTHead, a framework for point-based neural rendering from monocular images
  • Conducted comprehensive benchmarks against leading methods for cross-identity reenactment
  • Demonstrated state-of-the-art approaches on VoxCeleb1 and VoxCeleb2, while additionally improving efficiency (Accepted by WACV2024)

    • Trajectory Prediction and Driving Video Caption

    Assistant Prof. Hao Zhao, THU
    NumPy, PyTorch
  • predicted trajectory on a new interactive motion dataset by AgentFormer and Trajectron++
  • trained a novel end-to-end transformer generated description and explanation of the video
  • demonstrated state-of-the-art performance (a co-authored paper accepted to ICRA 2023)

    • Capstone Projects

    Inference Acceleration for Face Detection Models

    C/C++ Programming and Design (CS205), SUSTech
    C++, AVX
  • re-implemented convolutional layer, fully connected layer and pooling layer by C++ and AVX
  • improved the inference speed by 60 times via replacement of operators

    • A Local Search Heuristic for the Capacitated Arc Routing Problem (CARP)

    Artificial Intelligence (CS303), SUSTech
    Python
  • proposed a new local search algorithm for CARP and accelerated it by multiprocessing
  • achieved the best performance in the online judge among more than 150 students

    • Professional Experience

    Lightweight OCR Models for OpenCV

    OpenCV@Google Summer of Code
    PyTorch, ONNX, C++
  • implemented the detection part of PP-OCRv2 model in OpenCV Zoo by ONNX
  • implemented high level C++ API of PP-OCRv2 model in OpenCV
  • implemented evaluation metrics of text detection (AP, Recall, Precision, Hmean) in OpenCV Zoo

    • Undergraduate Teaching Assistant

    CS102B, SUSTech
    Java
  • designed and graded half of assignments, final exam and project

    • Publications

  • Tong Zhang, Yiming Chen, Simin Chen, Zexin Li, Xianghu Yue, Cong Liu, Chenyu You, and Haizhou Li, "Unintended Side Effects of Defense Mechanisms in Large Language Models: A Comprehensive Study", Under Review, 2025.
  • Tong Zhang, Haoyang Liu, Peiyan Zhang, Yuxuan Cheng, and Haohan Wang, "Beyond Pixels: Exploring Human-Readable SVG Generation for Simple Images with Vision Language Models", Preprint, 2024.
  • Haoyu Ma, Tong Zhang, Shanlin Sun, Xiangyi Yan, Kun Han, and Xiaohui Xie, "CVTHead: One-shot Controllable Head Avatar with Vertex-feature Transformer", IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024.
  • Bu Jin, Xinyu Liu, Yupeng Zheng, Pengfei Li, Hao Zhao, Tong Zhang, Yuhang Zheng, Guyue Zhou and Jingjing Liu, "ADAPT: Action-aware Driving Caption Transformer", IEEE International Conference on Robotics and Automation (ICRA), 2023.

    • Facts about Me

  • Hometown: Wuhan
  • Idol: Richard Feynman
  • Dream: To be a great researcher and design influential software
  • I enjoy finding potential topics from active discussions and am proud of my creativity.
    • This page has been accessed several times since Jan. 10, 2023.