Di Un Pak

About

Di Un Pak is a rising researcher in computer vision and autonomous driving, whose work focuses on bridging the domain gap in 3D perception. His most-cited paper, "DAPS3D: Domain Adaptive Projective Segmentation of 3D LiDAR Point Clouds" (2023, 13 citations), tackles a critical bottleneck for self-driving vehicles: the inability of segmentation models trained on synthetic data to generalize to real-world LiDAR scans. By developing a domain-adaptive projective framework, Pak enables neural networks to reliably segment roads, buildings, pedestrians, and vehicles across different environments without costly manual re-annotation. This contribution directly addresses the limitations of point- and voxel-based segmentation architectures, which often fail under varying weather, sensor configurations, or geographic conditions. Though early in his career, Pak’s work has already attracted attention for its practical relevance to autonomous navigation and robotics. His research represents a promising step toward robust, scalable perception systems that can operate safely in the wild.

Research Focus

Key Achievements

Top Papers

1. 1
   DAPS3D: Domain Adaptive Projective Segmentation of 3D LiDAR Point Clouds

   13 citations · 2023

Key Collaborators

Contact & Links