RelateGS: Reliable Semantic and Relational Evidence for 3D Scene Understanding with Gaussian Splatting

WACV 2027

Seongjun Choi · Sangyoun Lee^†

Yonsei University · School of Electrical and Electronic Engineering

^†Corresponding author

TL;DR RelateGS improves relation-aware 3D visual grounding by selectively refining the scene-side evidence of a language-aware Gaussian representation. It suppresses unreliable multi-view semantic observations and uses a persistent 3D object graph to identify and recover under-observed object relations from informative existing views.

📢 News

• 2026.10 RelateGS is accepted to WACV 2027! See you in Orlando.
• Code release: Implementation and model artifacts will be released progressively. Watch this repository for updates.

🗂️ Release Plan

• Inference code
• Pretrained checkpoints
• Training code
• Preprocessing data (relation annotation)

📖 Citation

@inproceedings{choi2027relategs,
  title     = {RelateGS: Reliable Semantic and Relational Evidence for 3D Scene Understanding with Gaussian Splatting},
  author    = {Choi, Seongjun and Lee, Sangyoun},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  year      = {2027}
}