University of North Carolina, Chapel Hill
— Ph.D. student, Department of Computer Science
2024 – Present · Chapel Hill, NC
My research focuses on developing robust and efficient frameworks for computational photography and scientific imaging with GenAI. In particular, I work on high-speed, high-quality 3D imaging through multi-modal sensor fusion and optics design. I also have broad interests in imaging inverse problems, generative AI, and metalens design.
Keywords: Computer Vision, Computational Imaging, Large-scale Optimization, Machine Learning, Event Camera, TOF Camera, Camera Fusion, Diffusion Models for Imaging, Metalens Design.
Developed a diffusion transformer with probabilistic temporal encoding to fuse asynchronous RGB and event streams for temporally consistent novel-view synthesis.
Investigated how visual scene properties bias perceived speed and developed a predictive model validated in immersive VR environments.
Explored implicit and explicit volumetric representations for integrating complementary sensor measurements.
Fuse RGB, depth, and event cameras together under a unified framework, enabling high-speed and small-baseline 3D/4D imaging and inverse rendering.
Combined RGB camera and event camera with special optical designs, enabling high-speed, high-resolution, and high-dynamic-range light-field imaging.
Explored physical model-based video diffusion models for imaging inverse problems, such as denoising, deblurring, and spatial-temporal interpolation, with applications to face video restoration.
Applied structured pruning and fine-tuning to accelerate model-based image reconstruction with minimal loss in reconstruction quality.
Proposed “ELDER”, a deep equilibrium regularizer that learns explicit regularization functionals, combining the effectiveness of deep learning with the convergence guarantees of optimization theory.
Investigated the robustness of deep equilibrium models for imaging inverse problems under physical model changes.