Gaussian splatting viewer written in Vulkan.

The main goal of this project is to maximize rendering speed.

For more details, refer to details.

Currently I am not actively maintaining this project, but I am still open to suggestions or discussions, and will be happy to answer any questions or make quick fixes. Please open github issues about anything. Job offer is also welcome ;)

Consider using vk_gaussian_splatting, another Vulkan 3DGS renderer by NVidia with richer functionalities. They mentioned my project (which is a great honor for me), and obviously their codes are far more well-organized.

Also, check out vulkan_radix_sort. It has a way better user-friendly interface that can be easily used by other projects. It is adopted by the NVidia sample, implying that the sort algorithm is quite stable.

Please check out splatstream for more organized c++ codes and experimental version of python binding.

Viewer works with pre-trained vanilla 3DGS models as input.

• Fast rendering speed
  • 350+ FPS on 1600x900, high-end GPU (NVidia GeForce RTX 4090)
  • 50+ FPS on 1600x900, high-end MacOS laptop (Apple M2 Pro)
  • 1-1.5x speed compared to SIBR viewer, but difference becomes bigger when scene is zoomed out,
    • because the number of tiles increases, and
    • more splats overlap in a single tile, so sequential blending operation takes more time
• Using graphics pipeline
  • Draw gaussian splats over other opaque objects, interacting with depth buffer
• 100% GPU tasks
  • No CPU-GPU synchronization for single frame: while GPU is working on frame i, CPU prepares a commands buffer and submits for frame i+1. No synchronization for frame i to get number of visible splats.
  • Indirect sort & draw: sorting and rendering only visible points
  • My vulkan radix sort implementation

• VulkanSDK>=1.3.296.0
  • slangc (or slangc.exe for Windows) is included in or above 1.3.296.0.
  • Required by vulkan_radix_sort submodule to compile slang shader files.
  • Download the latest version from https://vulkan.lunarg.com/ and follow install instruction.
• cmake>=3.15

• submodules

  $ git submodule update --init --recursive

  • VulkanMemoryAllocator
  • glm
  • glfw
  • imgui
  • argparse
  • vulkan_radix_sort: my Vulkan/GLSL implementation of reduce-then-scan radix sort.

$ cmake . -B build
$ cmake --build build --config Release -j

$ ./build/vkgs_viewer  # or ./build/Release/vkgs_viewer
$ ./build/vkgs_viewer -i <ply_filepath>

Drag and drop pretrained .ply file from official gaussian splatting, Pre-trained Models (14 GB).

• Left drag to rotate.

• Right drag, or Ctrl + left drag to translate.

• Left+Right drag to zoom in/out.

• WASD, Space to move.

• Wheel to zoom in/out.

• Ctrl+wheel to change FOV.

GUI is created in an off thread. According to GLFW documentation, the user should create window in main thread. However, managing windows off-thread seems working in Windows and Linux somehow.

Unfortunately, MacOS doesn't allow this. MacOS’s UI frameworks can only be called from the main thread. Here's a related thread by Apple staff.

• Windows or Linux (Doesn't work for MacOS.)
• conda: cmake, pybind11, cuda-toolkit (cuda WIP, not necessary yet)

$ conda create -n pygs python=3.10
$ conda activate pygs
$ conda install conda-forge::cmake
$ conda install conda-forge::pybind11
$ conda install nvidia/label/cuda-12.2.2::cuda-toolkit  # or any other version

The python package dynamically links to c++ shared library file.

So, first build the shared library first, then install python package.

$ cmake . -B build
$ cmake --build build --config Release -j
$ pip install -e binding/python

$ python
>>> import pygs
>>> pygs.show()
>>> pygs.load("./models/bicycle_30000.ply")  # asynchronously load model to viewer
>>> pygs.load("./models/garden_30000.ply")
>>> pygs.close()