Welcome to the StochasticSplats project repository!
Accepted to ICCV 2025
Shakiba Kheradmand1,2,*, Delio Vicini3,*, George Kopanas3,4,†, Dmitry Lagun1, Kwang Moo Yi2, Mark Matthews1, Andrea Tagliasacchi1,5,6
1Google DeepMind, 2University of British Columbia, 3Google, 4Runway ML, 5Simon Fraser University, 6University of Toronto
This repository builds upon and extends the capabilities of Splatapult, a 3D Gaussian splats viewer written by Anthony J. Thibault. StochasticSplats introduces stochastic rasterization methods enabling sorting-free 3D Gaussian splatting. You can run the app on desktop or in VR mode.
For a detailed explanation of controls and features, please refer to the original README.
Found a bug or have an idea to improve this project?
I’d love your input! Please open a pull request.
Thank you for helping make this project better.
To run StochasticSplats, follow these steps:
Ensure your scene is in the standard 3D Gaussian Splatting format, which consists of a directory containing:
cameras.jsoninput.ply- A
pointcloud/subdirectory with the point cloud data.
Execute the viewer from your build directory, providing the path to your scene.
Command Syntax:
splatapult.exe [options] [path/to/scene] [options]
./build/splatapult [path/to/scene] [options]Example on Desktop:
splatapult.exe path/to/my/scene --render_mode [AB | ST | ST-popfree] --width 1920 --height 1080Example on VR:
splatapult.exe -v path/to/my/scene --render_mode [AB | ST | ST-popfree] --width 1692 --height 1824The following table details the command-line arguments added in this code base. For the complete options, please refere to original README.
| Option | Description | Default |
|---|---|---|
--width |
Sets the width of the application window. | 1296 |
--height |
Sets the height of the application window. | 840 |
--render_mode |
Specifies the rendering mode.
|
AB |
--samples |
Defines the number of samples for stochastic modes. The maximum value depends on your hardware. | 1 |
--no-taa |
Disables Temporal Anti-Aliasing (TAA). By default, TAA is enabled but automatically turns off when samples > 1. | false |
If you utilize StochasticSplats in your research, please cite our ICCV 2025 paper:
@inproceedings{kv2025stochasticsplats,
title={StochasticSplats: Stochastic Rasterization for Sorting-Free 3D Gaussian Splatting},
author={Kheradmand, Shakiba and Vicini, Delio and Kopanas, George and Lagun, Dmitry and Yi, Kwang Moo and Matthews, Mark and Tagliasacchi, Andrea},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
year={2025}
}