GLINT: Modeling Scene-Scale Transparency via Gaussian Radiance Transport

CVPR 2026

Youngju Na1,2,* Jaeseong Yun2 Soohyun Ryu2 Hyunsu Kim2 Sung-Eui Yoon1 Suyong Yeon2
KAIST 1
NAVER LABS 2
Paper arXiv Code Dataset

TL;DR

GLINT decomposes outgoing radiance into interface, transmission, and reflection components with Gaussian radiance transport, enabling scene-scale transparency reconstruction with physically grounded geometry.

Videos

Abstract

While 3D Gaussian splatting has emerged as a powerful paradigm, it fundamentally fails to model transparency such as glass panels. The core challenge lies in decoupling the intertwined radiance contributions from transparent interfaces and the transmitted geometry observed through the glass. We present GLINT, a framework that models scene-scale transparency through explicit decomposed Gaussian representation. GLINT reconstructs the primary interface and models reflected and transmitted radiance separately, enabling consistent radiance transport. During optimization, GLINT bootstraps transparency localization from geometry-separation cues induced by the decomposition, together with geometry and material priors from a pre-trained video relighting model. Extensive experiments demonstrate consistent improvements over prior methods for reconstructing complex transparent scenes.

GLINT teaser showing rendered image, normal map, transparency map, decomposed radiance components, and reconstructed mesh.
GLINT reconstructs transparent scenes by explicitly separating interface, transmission, and reflection components, enabling physically grounded geometry and appearance.

Pipeline

GLINT method overview showing interface, transmission, and reflection Gaussian components and the rendering pipeline.
Overview of the GLINT pipeline. The method decomposes Gaussian radiance transport into interface, transmission, and reflection paths and supervises the output with photometric and geometric guidance.

Surface Reconstruction

GLINT surface reconstruction mesh for scene 1.
Baseline surface reconstruction mesh for scene 1.
TSGS
Ours
Surface reconstruction comparison for scene_1: GLINT vs. selected baseline.

Qualitative Results

The paper reports that GLINT improves both appearance and geometry on synthetic 3D-FRONT-T scenes and on real-world DL3DV-10K scenes, while producing more coherent transparent interfaces and fewer floating artifacts than prior Gaussian-based baselines.

Additional qualitative result for GLINT on a transparent scene.
Second qualitative result for GLINT on a transparent scene.
Detail crops and geometry outputs from GLINT qualitative results.
Transparency map visualizations for GLINT.

Qualitative Comparisons

Direct video comparison between GLINT and EnvGS on a representative transparent scene.

Ours vs. EnvGS qualitative comparison.

Citation

@inproceedings{na2026glint,
  title     = {GLINT: Modeling Scene-Scale Transparency via Gaussian Radiance Transport},
  author    = {Na, Youngju and Yoon, Jaeseong and Ryu, Soohyun and Kim, Hyunsu and Yoon, Sung-Eui and Yeon, Suyong},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026}
}