LangSplatV2: High-dimensional 3D Language Gaussian Splatting with 450+ FPS

Wanhua Li*, Yujie Zhao*, Minghan Qin*, Yang Liu, Yuanhao Cai, Chuang Gan, Hanspeter Pfister
(* indicates equal contribution)

This repository contains the official authors implementation associated with the paper "High-dimensional 3D Language Gaussian Splatting with 450+ FPS". We further provide the datasets and model weights.

Clone the Repository

The repository contains submodules, thus please check it out with

# SSH
git clone git@github.com:ZhaoYujie2002/LangSplat-V2.git --recursive

or

# HTTPS
git clone https://github.com/ZhaoYujie2002/LangSplat-V2.git --recursive

Setup

Requirements

• Conda (recommended for easy setup)
• C++ Compiler for PyTorch extensions (we used VS Code)
• CUDA SDK 11 for PyTorch extensions (we used 11.8)
• C++ Compiler and CUDA SDK must be compatible
• A NVIDIA A100 GPU (we used in our experiments)

Environment Setup

Our default, provided install method is based on Conda package and environment management:

conda env create --file environment.yml
conda activate langsplat_v2

Prepare Dataset

In the experiments section of our paper, we primarily utilized Three datasets: the LERF dataset and the 3D-OVS dataset and the Mip-NeRF360 dataset.

The LERF dataset is accessible for download from repository LangSplat via the following link: Download LERF Dataset.

The 3D-OVS dataset is accessible for download from repository 3D-OVS via the following link: Download 3D-OVS Dataset.

The Mip-NeRF360 dataset is accessible for download from repostory GAGS via the following link: Download Mip-NeRF360 Dataset.

For your own scenes, you need to acquire the following dataset format and a pre-trained RGB model follow the 3D Gaussian Splatting repository.

<dataset_name>
|---images
|   |---<image 0>
|   |---<image 1>
|   |---...
|---output
|   |---<dataset_name>
|   |   |---point_cloud/iteration_30000/point_cloud.ply
|   |   |---cameras.json
|   |   |---cfg_args
|   |   |---chkpnt30000.pth
|   |   |---input.ply
|---sparse
    |---0
        |---cameras.bin
        |---images.bin
        |---points3D.bin

QuickStart

Download the pretrained model to output/, then simply use

# For the LERF dataset
# eg. bash eval_lerf.sh teatime 0 10000
bash eval_lerf.sh $scene_name $model_idx $checkpoint

Pipeline

The pipeline for training LangSplat V2 and evaluation.

• Step 1: Generate Language Feature of the Scenes.

  python preprocess.py --dataset_path $dataset_path 

• Step 2: Train the Global Semantic Codebook and the Sparse Coefficient Field.

  bash train.sh $dataset_root_path $scene_name $model_idx

• Step 3: Eval.

  For the LERF dataset

  bash eval_lerf.sh $scene_name $model_idx $checkpoint

TODO list:

• update the arxiv link
• release the preprocessed dataset and the pretrained model
• release more code of evaluation

BibTeX

@misc{li2025langsplatv2highdimensional3dlanguage,
      title={LangSplatV2: High-dimensional 3D Language Gaussian Splatting with 450+ FPS}, 
      author={Wanhua Li and Yujie Zhao and Minghan Qin and Yang Liu and Yuanhao Cai and Chuang Gan and Hanspeter Pfister},
      year={2025},
      eprint={2507.07136},
      archivePrefix={arXiv},
      primaryClass={cs.GR},
      url={https://arxiv.org/abs/2507.07136}, 
}