Related Experiment Video
Updated: Jan 9, 2026

Automated 3D Optical Coherence Tomography to Elucidate Biofilm Morphogenesis Over Large Spatial Scales
Published on: August 21, 2019
Gaussian Splatting with a Shading Model for Endoscope Image Visualization
Abstract:
Gaussian splatting is a technique to represent 3D object models by means of clouds of Gaussians optimized for multi-view images, enabling free view synthesis. This method assumes static illumination, but this assumption is violated when applied to the endoscopy images, where the light source is attached to the camera. As a result, this can produces floating artifact Gaussians. To address this issue, we propose a method that applies the shader to the output image of the Gaussian renderer. This shader incorporates diffuse and specular reflection models and vignetting. It is optimized with the Gaussian model to ensure that the shader's output image closely resembles the ground truth input image. Our method was tested on both synthetic and real images, demonstrating its effectiveness.

