A research team led by Professor In-Kwon Lee at Yonsei University (Kyumin Kim, Dohae Lee, Hanul Back, and In-Kwon Lee) has achieved a significant milestone with their paper, "RotGS: Rotation-Guided 3D Gaussian Splatting for Turntable Sequences without Structure-from-Motion." The study has been accepted for presentation at Eurographics 2026 and will be simultaneously published in the prestigious Computer Graphics Forum (CGF) journal.
The study tackles a critical bottleneck in 3D reconstruction. While 3D Gaussian Splatting (3DGS) has revolutionized photorealistic scene representation, it traditionally relies on Structure-from-Motion (SfM) to determine precise camera posesโa process that is often unstable in turntable setups. In these environments, background removal reduces reliable feature matches, leading to inaccurate poses and degraded reconstruction quality.
To solve this, the team introduced RotGS, a novel SfM-free method optimized for turntable sequences. By leveraging the geometric prior that objects rotate around a central axis, RotGS simplifies the optimization process using a single global rotation axis.
The work will be presented at the 47th Annual Conference of the European Association for Computer Graphics (Eurographics 2026) in Bonn, Germany. This breakthrough is expected to significantly streamline automated 3D content creation for e-commerce, digital twins, and cultural heritage preservation.