InternLandMark/FlashGS

FlashGS

FlashGS is an efficient CUDA Python library, enabling real-time 3D Gaussian Splatting (3DGS) based rendering especially for large-scale and high-resolution (4K or even higher) scenes.

[Project Page]

Hardware Requirement

NVIDIA's server-grade and consumer-grade GPUs should work for our implementation. We have conducted our experiments on an NVIDIA A100, V100, RTX 2080ti, RTX 3090 and RTX 4090 GPUs.

Directories

• csrc/: Our CUDA C++ implementation of FlashGS. The optimized rendering kernels are under csrc/cuda_rasterizer/.
• example.py: An example to show how to use the installed FlashGS library.
• setup.py: A Python script to build, package, and install the FlashGS library.
• requirements.txt: Record some software dependencies when installing FlashGS.

Installation

You can follow the following steps to setup on your machine:

• Clone the FlashGS project from this page.
• Download the dependencies as we recommend.
• Use python setup.py install or pip install . to install FlashGS library.
• Run pip uninstall flash-gaussian-splatting before you compile and install the new version.

Run Example

• Download the pre-trained models. https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/datasets/pretrained/models.zip
• Run python example.py model_path.
• Open model_path/test_out and check the result.