Our code is based on 3D Gaussian Splatting.
Examples of factorized coordinates: (a) 25 normal coordinates, (b) 5 × 5 factorized coordinates. each x and y axis has 5 points, and both represent 25 (5 × 5) points. (c) two 5 × 5 factorized coordinates and a total of 50 points are represented (2 × 5 × 5), (d) multi-resolution factorized coordinates, where two factorized coordinates have different resolutions (3 × 3 and 5 × 5), represent total 34 points, (e) two 3 × 3 and one 5 × 5 factorized coordinates. A total of 43 points are represented. The best-viewed in color.
For installation, we use the same environment as 3D Gaussian Splatting.
We use the same environment as original 3DGS, please follow the link of 3D-GS to install all packages.
https://github.com/graphdeco-inria/gaussian-splatting
Thanks for the excellent work in 3D-GS!
Here, we will explain how to use our codes. Our codes are divided into two parts, one for synthetic-nerf dataset and other for Tanks&Temples dataset.
To use our code, first we need to use original 3DGS code to train each scene and get the .ply file for each scene. The distribution of original Gaussians will be used in next step.
python train.py -s /workspace/datasets/nerf_synthetic/chair -m exp/chair --eval --hist_path /gaussian-ori/gaussian-splatting/exp/chair/point_cloud/iteration_30000/point_cloud.ply
the source of dataset
the path of .ply file trained by original 3DGS
the output of the model
python train.py -s /workspace/datasets/TanksAndTemple/Barn -m TanksAndTemple/Barn --eval -r 2 -w --hist_path /gaussian-ori/gaussian-splatting/TanksAndTemple/Barn/point_cloud/iteration_30000/point_cloud.ply
the source of dataset
the path of .ply file trained by original 3DGS
the output of the model
the resolution of images
the background is white
Refer to other arguments of 3DGS.
@article{sun2024f,
title={F-3DGS: Factorized Coordinates and Representations for 3D Gaussian Splatting},
author={Sun, Xiangyu and Lee, Joo Chan and Rho, Daniel and Ko, Jong Hwan and Ali, Usman and Park, Eunbyung},
journal={arXiv preprint arXiv:2405.17083},
year={2024}
}