Official Pytorch implementation "Identity-aware Language Gaussian Splatting for Open-vocabulary 3D Semantic Segmentation"
SungMin Jang, and Wonjun Kim (Corresponding Author)
🌸IEEE/CVF International Conference on Computer Vision, (ICCV), Oct. 2025.🌸
[ Training pipeline ]
We propose Identity-aware Language Gaussian Splatting (ILGS), a simple yet powerful method to resolve view-inconsistent language embeddings for open-vocabulary 3D semantic segmentation.
ILGS introduces two key components: an identity-aware semantic consistency loss to ensure consistent features across views, and a progressive mask expanding scheme for precise boundary segmentation.
We provide:
- ✅ Full Implementation: The complete source code for ILGS.
- 🪄 3D Editing: Examples of object removal, color modification, and resizing.
For detailed instructions on how to set up the environment and install dependencies, please refer to our Installation.md
We provide guidelines to download datasets.
Please check Download.md for more information.
To get started with the full implementation of our open-vocabulary segmentation model, please follow the instructions in the Implementation.md.
For instructions on our 3D editing features, please refer to the Editing.md. This guide explains how to launch the interactive demos and use functionalities like object removal, color modification, and resizing.
Below are the semantic segmentation results of our proposed method on the LERF-Mask benchmark.
This project is licensed under the Apache License 2.0, with the exception of certain components derived from the Gaussian Splatting project.
- Apache License 2.0: All original code written for ILGS is released under the Apache 2.0 license. See LICENSE.
- Non-commercial License (Inria & MPII): Some parts of the code are based on Gaussian Splatting, which is licensed for non-commercial research use only. See LICENSE_GAUSSIAN_SPLATTING.md for full terms.
Please ensure that you comply with both licenses when using this repository.
This work was supported by the National Research Foundation of Korea (NRF) funded by the Korea Government (MSIT) under Grant RS-2023-NR076462.
Our implementation and experiments are built on top of open-source GitHub repositories. We thank all the authors who made their code public, which tremendously accelerates our project progress. If you find these works helpful, please consider citing them as well.
lkeab/gaussian-grouping
minghanqin/LangSplat
If you find our work useful for your project, please consider citing the following paper.
@inproceedings{jsmbankILGS,
title={Identity-aware Language Gaussian Splatting for Open-vocabulary 3D Semantic Segmentation},
author={SungMin Jang and Wonjun Kim},
booktitle={Proceedings of the International Conference on Computer Vision},
year={2025}
}