Hello, I'm Hyoungseob Park

Research Interest

My research interest is utilizing heterogeneous and complementary multimodal sensory information, which sits at the intersection of 3D reconstruction, multimodal sensor fusion, multimodal representation learning, and adaptive vision, in the context of autonomous driving and robotic perception. In Summer 2025, I joined NVIDIA as a Ph.D. research intern (May 2025 - Aug 2025), focusing on the problem of grounding the dynamic obstacle for autonomous driving by utilizing a 3D Vision-Language Model. Before that, I joined Amazon AGI (June 2024 - Sep 2024), working on audio-visual model compression.

Reconstructing the 3D scene from a single RGB image is an ill-posed problem. My research focuses on metric-scale 3D scene reconstruction and Computer Vision with multimodal complementary sensory information as a condition.

(1) Depth Completion: ProxyTTA (CVPR’24), AllDayDepth (IROS’24), ProtoDepth (CVPR’25), ETA (ICCV’25), ORCaS (ICLR’26)

(2) Robust and adaptive (3D) Vision: ProxyTTA (CVPR’24), ETA (ICCV’25), ProgressiveTTA (ICCV’25)

(3) Non-Perspective Camera Calibration: Calibration Token (ICCV’25)

(4) Heterogeneous Modality Fusion: UniTouch (CVPR’24), KTD (ICLR’26)

Research Highlights