I am a postdoctoral researcher at VILAB, Hanyang University, advised by Prof. Tae Hyun Kim, with whom I also completed my Ph.D. I received my B.S. in Mechanical Engineering from the same university. My research focuses on restoring and enhancing images and videos degraded by sub-optimal capture conditions, with the broader goal of turning any capture into a good photo. I approach this problem from three directions: designing task-specific network architectures, formulating restoration as an inverse problem, and leveraging generative models.
What makes a photo good, however, depends on who is looking at it. While much of my work has focused on visual quality for human viewers, I am increasingly interested in enhancement for machines: improving images so that downstream vision models can perceive and act on them more reliably. If you are interested in collaboration, feel free to contact me.
Recovers sharp frames at arbitrary continuous time instants from a single blurry image by jointly learning deblurring and blur-to-video generation.
A lightweight network that performs uncertainty-aware adaptive processing to restore high-frequency details in regions with varying UDC degradations.
A prompt-based slot-in-slot attention mechanism that clusters exposed feature regions to learn exposure-centric features.
Formulates low-light enhancement as a continuous system with Neural ODEs, enabling stable curve-based exposure correction without sensitivity to iteration count.
Maps features to diverse distributions based on semantic location, improving discrimination between normal and abnormal patterns.
* Equal contribution