SpeclatentGS

This repo contains the official implementation of the ACM MM 2024 paper :

SpecGaussian with latent features: A high-quality modeling of the view-dependent appearance for 3D Gaussian Splatting

Zhiru Wang，Shiyun Xie，Chengwei Pan and Guoping Wang

PIPELINE

Environment Installation

You can install the base environment using:

git clone https://github.com/MarcWangzhiru/SpeclatentGS.git
cd SpeclatentGS
conda env create --file environment.yml

For the installation of submodules, you can use the following command:

cd submodules/diff-gaussian-rasterization
python stup.py install

and

cd submodules/simple-knn
python stup.py install

You also need to install the tinycudann library. In general, you can use the following command:

pip install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

Dataset Preparation

The dataset used in our method is in the same format as the dataset in Gaussian Splatting. If you want to use your custom dataset, follow the process of in Gaussian Splatting. We obtained our own shiny_dataset by resize the images of original Shiny Dataset and recolmap.

Trainng

For training, you can use the following command:

python train.py -s <path to COLMAP or NeRF Synthetic dataset> --eval

Evalution

For evalution, you can use the following command:

python render.py -m <path to trained model> --eval

Citation