4D Triangle Splatting Reconstruction of Dynamic Endoscopic Scenes from Monocular Videos

This paper introduces SurgicalTS, a self-supervised triangle-splatting framework designed for robust 4D reconstruction of deformable endoscopic scenes using static monocular cameras. The method employs a canonical set of triangles jointly optimized with a deformation network, incorporating quaternion-based deformation modeling, improved color parametrization, and spatio-temporal priors to handle non-rigid tissue deformation, illumination changes, and occlusions. Extensive evaluation on in-vivo endoscopy videos demonstrates that SurgicalTS effectively models soft-tissue deformations and tool-tissue interactions over time, outperforming state-of-the-art methods while reducing training times and maintaining competitive geometric and photometric accuracy.