Voxel51/gaussian_splatting · Datasets at Hugging Face

Gaussian Splats Dataset

3D Gaussian Splatting for Real-Time Radiance Field Rendering

Dataset Author: Paula Ramos
Created Using: 3D Gaussian Splatting Paper
Code Repository: GitHub - graphdeco-inria/gaussian-splatting

Description

This dataset consists of Gaussian Splats representations of different real-world scenes, created using the official 3D Gaussian Splatting method. Each scene folder contains:

A reference image representing the scene.
A PLY file stored in a point_cloud_folder, containing the Gaussian Splats reconstruction.

Overview

This dataset consists of Gaussian Splats representations of different real-world scenes, created using the official 3D Gaussian Splatting method. Each scene folder contains:

A reference image representing the scene.
Two PLY files stored in a point_cloud_folder, containing the Gaussian Splats reconstructions at iterations 7000 and 30000.

The dataset is structured as follows:

FO_dataset/
│── drjohnson/          # Scene Folder
│   ├── reference_image.png
│   ├── point_cloud_folder/
│       ├── reconstruction_7000.ply
│       └── reconstruction_30000.ply
│── playroom/
│   ├── reference_image.png
│   ├── point_cloud_folder/
│       ├── reconstruction_7000.ply
│       └── reconstruction_30000.ply
│── train/
│   ├── reference_image.png
│   ├── point_cloud_folder/
│       ├── reconstruction_7000.ply
│       └── reconstruction_30000.ply
│── truck/
│   ├── reference_image.png
│   ├── point_cloud_folder/
│       ├── reconstruction_7000.ply
│       └── reconstruction_30000.ply

How to Use the Dataset

1. Install the Required FiftyOne Plugin

To visualize all .ply files using FiftyOne, download the Gaussian Splats plugin:

!fiftyone plugins download https://github.com/danielgural/ksplats_panel

2. Load & Visualize the Dataset with FiftyOne

Use the following Python script to load and explore the dataset in FiftyOne:

import fiftyone as fo
from fiftyone.utils.splats import SplatFile

# Create a FiftyOne dataset
dataset = fo.Dataset(name="splat-test", overwrite=True)

# Add samples (update paths as needed)
sample1 = fo.Sample(filepath="FO_dataset/drjohnson/reference_image.png")
sample1["splat"] = SplatFile(filepath="FO_dataset/drjohnson/point_cloud_folder/reconstruction_30000.ply")

sample2 = fo.Sample(filepath="FO_dataset/drjohnson/reference_image.png")
sample2["splat"] = SplatFile(filepath="FO_dataset/drjohnson/point_cloud_folder/reconstruction_7000.ply")

sample3 = fo.Sample(filepath="FO_dataset/playroom/reference_image.png")
sample3["splat"] = SplatFile(filepath="FO_dataset/playroom/point_cloud_folder/reconstruction_7000.ply")

sample4 = fo.Sample(filepath="FO_dataset/playroom/reference_image.png")
sample4["splat"] = SplatFile(filepath="FO_dataset/playroom/point_cloud_folder/reconstruction_30000.ply")

sample5 = fo.Sample(filepath="FO_dataset/train/reference_image.png")
sample5["splat"] = SplatFile(filepath="FO_dataset/train/point_cloud_folder/reconstruction_7000.ply")

sample6 = fo.Sample(filepath="FO_dataset/train/reference_image.png")
sample6["splat"] = SplatFile(filepath="FO_dataset/train/point_cloud_folder/reconstruction_30000.ply")

sample7 = fo.Sample(filepath="FO_dataset/truck/reference_image.png")
sample7["splat"] = SplatFile(filepath="FO_dataset/truck/point_cloud_folder/reconstruction_7000.ply")

sample8 = fo.Sample(filepath="FO_dataset/truck/reference_image.png")
sample8["splat"] = SplatFile(filepath="FO_dataset/truck/point_cloud_folder/reconstruction_30000.ply")

# Add samples to the dataset
dataset.add_sample(sample1)
dataset.add_sample(sample2)
dataset.add_sample(sample3)
dataset.add_sample(sample4)
dataset.add_sample(sample5)
dataset.add_sample(sample6)
dataset.add_sample(sample7)
dataset.add_sample(sample8)

# Launch FiftyOne App
session = fo.launch_app(dataset, auto=False, port=5152)

Visualization Results

Below are sample screenshots showcasing the 3D Gaussian Splats reconstructions:

Drjohnson Scene

Playroom Scene

https://github.com/user-attachments/assets/1c3d3b6b-2b7b-4e93-8f5c-76a184f51260

Train Scene

https://github.com/user-attachments/assets/78ca63f5-1df9-4970-a50c-bfab0ee3615f

Truck Scene

Research & Applications

This dataset is useful for a variety of 3D vision and AI applications, including:

NeRF & Gaussian Splatting Benchmarking
3D Scene Understanding & Reconstruction
Multi-Modal AI (Images + 3D Point Clouds)
Real-Time 3D Rendering Research

Citation

If you use this dataset, please cite the original 3D Gaussian Splatting paper:

@article{kerbl2023gsplatting,
  title={3D Gaussian Splatting for Real-Time Radiance Field Rendering},
  author={Kerbl, Bernhard and Kopanas, Georgios and Leimkühler, Thomas and Drettakis, George},
  journal={arXiv preprint arXiv:2308.04079},
  year={2023}
}

Datasets:

Voxel51
/

gaussian_splatting