Sungyong Park

Sungyong Park

Ph.D. Student in Digital Media (Artificial Intelligence) at Soongsil University

Department of Digital Media, Soongsil University
Seoul, Republic of Korea

ejqdl at soongsil dot ac dot kr   /   ejqdl010 at gmail dot com

About

I am a Ph.D. student at the Reality Lab in the Department of Digital Media at Soongsil University, advised by Prof. Heewon Kim.

My research focuses on computer vision and embodied AI, especially robust visual perception under real-world degradations and generalizable robot manipulation. I am interested in building AI systems that can perceive reliably in imperfect sensing conditions and act effectively in continuous, language-conditioned environments.

Research Interests

My research centers on robust perception and embodied intelligence for real-world AI systems. Key areas of interest include:

  • Embodied AI and Robotic Manipulation: Developing vision-language-action models and robot learning methods that generalize across scenes, objects, and continuous goal states.
  • Robust Visual Perception: Studying how real-world camera degradations, such as contaminated lenses, affect high-level perception tasks including semantic segmentation.
  • Image Restoration in Real-World Settings: Building datasets and methods for restoring images captured under realistic sensing conditions, including dirty lenses and cross-sensor noise.
  • Scalable Data Generation: Creating simulation and generative pipelines for producing diverse, physically meaningful data for embodied AI.

News

Publications

    (* indicates equal contribution)

  • CLP: A Real-World Dataset of Contaminated Lens Protectors for Robust Semantic Segmentation

    Sungyong Park*, Sooyoung Choi*, Hyunseo Koh, Youngjae Choi, and Heewon Kim, IEEE/CVF Computer Vision and Pattern Recognition Conference (CVPR), 2026

    PDF Website
    Robust semantic segmentation in real-world perception can be severely affected by contaminated lens protectors. CLP introduces a real-world dataset of contaminated lens protectors to evaluate and improve semantic segmentation robustness under realistic camera degradations.
    @inproceedings{park2026clp,
  title={CLP: A Real-World Dataset of Contaminated Lens Protectors for Robust Semantic Segmentation},
  author={Park, Sungyong and Choi, Sooyoung and Koh, Hyunseo and Choi, Youngjae and Kim, Heewon},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2026}
}

  • Toward Interpretable Space Image Denoising by Learning Cross-Sensor Celestial Signals

    Sungyong Park, Ji Hoon Kim, and Heewon Kim, NTIRE Workshop at CVPR, 2026

    PDF
    Space imagery often suffers from sensor-specific noise and degradation. This work studies interpretable space image denoising by learning cross-sensor celestial signals, aiming to separate reusable celestial structures from sensor-dependent noise for robust restoration.
    @inproceedings{park2026interpretable,
  title={Toward Interpretable Space Image Denoising by Learning Cross-Sensor Celestial Signals},
  author={Park, Sungyong and Kim, Ji Hoon and Kim, Heewon},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  year={2026}
}

  • DynScene: Scalable Generation of Dynamic Robotic Manipulation Scenes for Embodied AI

    Sangmin Lee*, Sungyong Park*, Heewon Kim, IEEE/CVF Computer Vision and Pattern Recognition Conference (CVPR), 2025

    Paper Slides
    Robotic manipulation in embodied AI critically depends on large-scale, high-quality datasets that reflect realistic object interactions and physical dynamics. We present DynScene, a diffusion-based framework for generating dynamic robotic manipulation scenes directly from textual instructions.
    @inproceedings{lee2025dynscene,
  title={DynScene: Scalable Generation of Dynamic Robotic Manipulation Scenes for Embodied AI},
  author={Lee, Sangmin and Park, Sungyong and Kim, Heewon},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={12166--12175},
  year={2025}
}

  • SIDL: A Real-World Dataset for Restoring Smartphone Images with Dirty Lenses

    Sooyoung Choi*, Sungyong Park*, Heewon Kim, AAAI Conference on Artificial Intelligence, 2025

    PDF Website Talk Slides
    Smartphone cameras are ubiquitous in daily life, yet their performance can be severely impacted by dirty lenses, leading to degraded image quality. SIDL introduces a real-world dataset for restoring images captured through contaminated smartphone lenses.
    @inproceedings{choi2025sidl,
  title={SIDL: A Real-World Dataset for Restoring Smartphone Images with Dirty Lenses},
  author={Choi, Sooyoung and Park, Sungyong and Kim, Heewon},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={3},
  pages={2545--2554},
  year={2025}
}

  • UDT: Unsupervised Discovery of Transformations between Fine-Grained Classes in Diffusion Models

    Youngjae Choi*, Hyunseo Koh*, Hojae Jeong*, Byungkwan Chae*, Sungyong Park, and Heewon Kim, British Machine Vision Conference (BMVC), 2025

    PDF
    This work studies unsupervised discovery of transformations between fine-grained classes in diffusion models, aiming to identify semantic directions that explain how one fine-grained class can be transformed into another without explicit paired supervision.
    @inproceedings{choi2025udt,
  title={UDT: Unsupervised Discovery of Transformations between Fine-Grained Classes in Diffusion Models},
  author={Choi, Youngjae and Koh, Hyunseo and Jeong, Hojae and Chae, Byungkwan and Park, Sungyong and Kim, Heewon},
  booktitle={British Machine Vision Conference},
  year={2025}
}

Awards

  • 2nd Place in the ARNOLD Challenge

    Sungyong Park, Heewon Kim
    CVPR 2026 Embodied AI Workshop

    Challenge Page Slides

  • 1st Place in the ARNOLD Challenge

    Dowon Kim, Chaewoo Lim, Sungyong Park, Sangmin Lee, Heewon Kim
    CVPR 2025 Embodied AI Workshop

    Challenge Page Slides

  • 3rd Place in the ARNOLD Challenge

    Sangmin Lee, Sungyong Park, Heewon Kim
    CVPR 2024 Embodied AI Workshop

    Challenge Page Slides