datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cl-splats-dataset
CL-Splats Dataset
This repository contains the data to reproduce our results from the CL-Splats paper. In addition, for future research we provide the Blender files used to create our synthetic scenes with a blender python script creating the camera trajectories.
Usage
The files under Blender-Levels and Real-World are compatible with the 3DGS data-loading. We will create a copy of this dataset that matches our data-loading script once available. Note that the COLMAP… See the full description on the dataset page: https://huggingface.co/datasets/ackermannj/cl-splats-dataset.gaussian_splatting
Gaussian Splats Dataset
3D Gaussian Splatting for Real-Time Radiance Field Rendering
Dataset Author: Paula RamosCreated Using: 3D Gaussian Splatting PaperCode 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… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/gaussian_splatting.Physics-Informed-Deformable-Gaussian-Splattingsplataverse-sketchfab
Splataverse: Sketchfab
splat-azur-001
Azurite Specimen (azur-001) — Gaussian Splat
Azurite specimen from Morenci, AZ (Phelps Dodge / Freeport-McMoRan mine) reconstructed using Gaussian Splatting. DSLR capture with dark matte background achieved excellent COLMAP registration (51/51 frames).
Specimen Details
Mineral: Azurite
Location: Morenci, AZ (Phelps Dodge / Freeport-McMoRan mine)
Listed on: Borussia Minerals
Capture Method
Device: DSLR
Total frames: 51 (crystal-cropped session)
Background:… See the full description on the dataset page: https://huggingface.co/datasets/idirectships/splat-azur-001.image-2-splat-DPO
Image-to-3D Gaussian Splatting DPO Dataset
A curated dataset of high-quality image-to-3D Gaussian Splatting pairs for Direct Preference Optimization (DPO) and contrastive learning, extracted from the Bittensor 404-gen-subnet competition.
Dataset Overview
This dataset contains 10,320 files (136GB+) organized into winner-loser pairs for training preference-based 3D generation models:
2,064 prompt images (original input images)
4,128 winner samples (2,064 PNG images + 2,064… See the full description on the dataset page: https://huggingface.co/datasets/Jyoti73ML/image-2-splat-DPO.eval_splatsim_approach_lever_benchmark_1000This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 1000,
"total_frames": 187307,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 20,
"splits": {
"train": "0:1000"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/eval_splatsim_approach_lever_benchmark_1000.cloth-splatting-folding-scenessplatsim_approach_lever_2_1path_stretch_baseThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 299,
"total_frames": 41874,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 50,
"splits": {
"train": "0:299"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": null… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_2_1path_stretch_base.Confident_SplattingConfident Splatting dataset is introduced in the paper Confident Splatting: Confidence-Based Compression of 3D Gaussian Splatting via Learnable Beta Distributions.
Project Page: https://amirhossein-razlighi.github.io/Confident-Splatting/
Code: https://github.com/AmirHossein-razlighi/Confident-Splatting
It is a dataset for 3D Reconstruction task, containing SFM pointcloud and different posed images. These different scenes have different challenges to reconstruct including floaters, distractor… See the full description on the dataset page: https://huggingface.co/datasets/AmirHossein-razlighi/Confident_Splatting.splatsim_approach_lever_11_50failsrrtpi05This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 380,
"total_frames": 88633,
"total_tasks": 2,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:380"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_11_50failsrrtpi05.1stpass10_splatsim_approach_lever_benchmark_1000_fastnotes to self: this is using the faster teleop (aka no velocity limiting)
This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 1012,
"total_frames": 188994,
"total_tasks": 2,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:1012"
},
"data_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/1stpass10_splatsim_approach_lever_benchmark_1000_fast.splatsim_approach_lever_12_cleanThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 300,
"total_frames": 70079,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:300"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_12_clean.splatsim_approach_lever_11_biasend_5pathThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 300,
"total_frames": 81392,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:300"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_11_biasend_5path.splat-amet-001
Amethyst Specimen (amet-001) — Gaussian Splat
Amethyst specimen reconstructed using Gaussian Splatting via COLMAP + opensplat pipeline on MacBook Pro.
Capture Method
Device: Camera/iPhone (handheld)
Total frames: 103 (images_jpg/ directory)
Background: Dark matte (controlled background for COLMAP registration)
COLMAP registration: Registered via colmap_ws_v1
Directory Structure
images_nobg/ — background-removed images (primary COLMAP input, symlinked as… See the full description on the dataset page: https://huggingface.co/datasets/idirectships/splat-amet-001.splat-cupr-001
Cuprite on Copper Specimen (cupr-001) — Gaussian Splat
Cuprite on native copper specimen from Ray Mine, AZ. Reconstructed using Gaussian Splatting after two capture sessions — the first (car seat background) failed COLMAP registration (4/25 frames); the second (dark matte background, 123 frames) achieved 115/123 registration.
Specimen Details
Mineral: Cuprite on native copper
Location: Ray Mine, AZ
Listed on: Borussia Minerals
Capture Sessions
Session… See the full description on the dataset page: https://huggingface.co/datasets/idirectships/splat-cupr-001.aquarium-gaussian-splatssplatsim_approach_lever_2_1path_wristThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 299,
"total_frames": 41874,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 50,
"splits": {
"train": "0:299"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": null… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_2_1path_wrist.splatmicrosplateval_splatsim_approach_lever_13_benchmarkThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.images.base_rgb_letterbox": {
"dtype": "image",
"shape": [
3,
224,
224
],
"names": [
"channels",
"height",
"width"
]
}… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/eval_splatsim_approach_lever_13_benchmark.splatsim_lerobot_splatsim_10fps_260609_0303This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 10,
"features": {
"observation.images.base_rgb": {
"dtype": "image",
"shape": [
3,
480,
640
],
"names": [
"channels",
"height",
"width"
]
},
"observation.images.wrist_rgb": {… See the full description on the dataset page: https://huggingface.co/datasets/LuEduSoHu/splatsim_lerobot_splatsim_10fps_260609_0303.splataverse-smithsonian
Splataverse: Smithsonian
splat-growth-artifactssplatsim_approach_lever_11_50failsrrtpi05_piabsden04This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 79,
"total_frames": 7024,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:79"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_11_50failsrrtpi05_piabsden04.splatsim_approach_lever_11_50failsrrtpi05_piabsden10This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 79,
"total_frames": 7024,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:79"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_11_50failsrrtpi05_piabsden10.splatssplatsim_approach_lever_11_50failsrrtpi05_piabsden06This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 79,
"total_frames": 7024,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:79"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_11_50failsrrtpi05_piabsden06.splatsim_approach_lever_11_50failsrrtpi05_piabsden02This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 79,
"total_frames": 7024,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:79"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_11_50failsrrtpi05_piabsden02.splatsim_approach_lever_2_1path_baseThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "lerobot_splatsim",
"total_episodes": 299,
"total_frames": 41874,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 50,
"splits": {
"train": "0:299"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": null… See the full description on the dataset page: https://huggingface.co/datasets/JennyWWW/splatsim_approach_lever_2_1path_base.