datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
PhysicalAI-WorldModel-Synthetic-Physical-Interaction-Scenes
PhysicalAI-WorldModel-Synthetic-Physical-Interaction-Scenes Dataset Card
Dataset Description
PhysicalAI-WorldModel-Synthetic-Physical-Interaction-Scenes is a large-scale synthetic dataset of physically-simulated multi-object interaction scenes, generated using NVIDIA Isaac Sim and the PhysX physics engine. It is designed to train and evaluate AI models on physical reasoning, rigid body dynamics, optical flow, depth estimation, and scene understanding.
Each clip… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/PhysicalAI-WorldModel-Synthetic-Physical-Interaction-Scenes.wds_imagenet_sketchimagenet1k-256-wdsThis is imagenet1k in webdataset format. Images are stored as jpg files. Every image has been resized to a maximum side length of 256. That means that if an image in the original dataset was 1000 by 500, the new size will be 256 by 128. Images with a maximum side length of under 256 were not resized.
The total size of all dataset files is 57.8 GB, there are 1,281,167 rows in the training split and 50,000 rows in the validation split.
pixelprose-shards
PixelProse Sharding Tars
arXiv | public-released version: pixelprose | JSON-only version: pixelprose-jsons
summary
Each tar file is approximately 500-600 MB, friendly for fast on-the-fly sampling, filtering, and loading in dataloaders.
Each tar file contains triplets of images, text, and JSON files. The *.txt files contain the raw original captions, while the *.json files include all the relevant information.
Due to Gemini-1.0 internal version changes during the… See the full description on the dataset page: https://huggingface.co/datasets/pixelprose/pixelprose-shards.pd12m-fullThis dataset is the downloaded variant of Spawning/PD12M. More specifically, this dataset
is compatible with webdataset. It was made public after obtaining permission
from the original authors of the dataset.
You can use the following to explore the dataset with webdataset:
import webdataset as wds
dataset_path = "pipe:curl -s -f -L https://huggingface.co/datasets/sayakpaul/pd12m-full/resolve/main/{00155..02480}.tar"
dataset = (
wds.WebDataset(dataset_path… See the full description on the dataset page: https://huggingface.co/datasets/Spawning/pd12m-full.3d_optical_flow_droid
3D Optical Flow DROID Dataset
Processed DROID robotics dataset with optical flow and scene flow annotations.
Dataset Structure
Organized by lab, each trajectory in separate tar.gz archive:
IPRL/IPRL+2023-06-19+Mon_Jun_19_23:27:48_2023.tar.gz
CLVR/CLVR+2023-...tar.gz
... (15 labs, ~33K trajectories)
Each trajectory contains:
metadata.json - Trajectory metadata
trajectory.h5 - Robot state and actions
camera_left/, camera_right/ - Camera data
rgb/ - RGB images
depth/ -… See the full description on the dataset page: https://huggingface.co/datasets/Salesforce/3d_optical_flow_droid.SpatialEdit-500K
SpatialEdit-500K
SpatialEdit-500K is a synthetic training dataset for fine-grained image spatial editing. It is built for learning geometry-aware edits such as object moving, object rotation, and camera viewpoint change.
The dataset was introduced in the paper SpatialEdit: Benchmarking Fine-Grained Image Spatial Editing. It is generated with a controllable rendering pipeline to provide structured spatial transformations at scale.
Project Resources
GitHub Repository:… See the full description on the dataset page: https://huggingface.co/datasets/EasonXiao-888/SpatialEdit-500K.BLIP3o-Pretrain-Short-Caption
BLIP3o Pretrain Short-Caption Dataset
This collection contains 5 million images, each paired with a short (~20 token) caption generated by Qwen/Qwen2.5-VL-7B-Instruct.
Download
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="BLIP3o/BLIP3o-Pretrain-Short-Caption",
repo_type="dataset"
)
Load Dataset without Extracting
You don’t need to unpack the .tar archives, use WebDataset support in 🤗datasets instead:
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/BLIP3o/BLIP3o-Pretrain-Short-Caption.minty-astro-ph
MINT-1T ArXiv Astro-ph
An astronomy-focused subset of mlfoundations/MINT-1T-ArXiv, filtered to include only papers from the astro-ph arXiv category (including cross-listed papers).
Overview
Papers
~845k
Total size
~804 GB
Format
WebDataset tar shards
Shards
287 (astro-ph-00000.tar to astro-ph-00286.tar)
Shard size
~3 GB each
Source
MINT-1T (Awadalla et al., 2024)
Data Format
Each tar shard contains paired files per paper:… See the full description on the dataset page: https://huggingface.co/datasets/Smith42/minty-astro-ph.pickapic_v2_webdatasetwebdataset archive of yuvalkirstain/pickapic_v2.
Dataloading code can be found here.
Stereo4D_vlbm
Stereo4D (converted to VLBM format)
This dataset contains 4,687 sequences from the Stereo4D dataset converted to the VLBM-compatible format using preprocess_stereo4d.py. The sequences have been compressed into .tar.gz archives in chunks of 50 sequences per archive.
Scale
Metric
Value
Total sequences
4,687
Image resolution
512 x 512 px
Depth type
Sparse (projected from tracked 3D points)
Dataset Structure
Each sequence directory follows this… See the full description on the dataset page: https://huggingface.co/datasets/ZhengGuangze/Stereo4D_vlbm.S1-MMAlignS1-MMAlign
A Large-Scale Multi-Disciplinary Scientific Multimodal Dataset
S1-MMAlign is a large-scale, multi-disciplinary multimodal dataset comprising over 15.5 million high-quality image-text pairs derived from 2.5 million open-access scientific papers.
Multimodal learning has revolutionized general domain tasks, yet its application in scientific discovery is hindered by the profound semantic gap between complex scientific imagery and sparse textual descriptions. S1-MMAlign aims to… See the full description on the dataset page: https://huggingface.co/datasets/ScienceOne-AI/S1-MMAlign.font-square-pretrain-20M
📚 Citation
If you use this dataset in your research, please cite these papers:
@article{pippi2023evaluating,
title={Evaluating Synthetic Pre-Training for Handwriting Processing Tasks},
author={Pippi, Vittorio and Cascianelli, Silvia and Baraldi, Lorenzo and Cucchiara, Rita},
journal={Pattern Recognition Letters},
year={2023},
publisher={Elsevier}
}
@InProceedings{pippi2025zeroshot,
author = {Pippi, Vittorio and Quattrini, Fabio and Cascianelli, Silvia and Tonioni… See the full description on the dataset page: https://huggingface.co/datasets/blowing-up-groundhogs/font-square-pretrain-20M.scannetppSemi-Truths
Semi Truths Dataset: A Large-Scale Dataset for Testing Robustness of AI-Generated Image Detectors (NeurIPS 2024 Track Datasets & Benchmarks Track)
Recent efforts have developed AI-generated image detectors claiming robustness against various augmentations, but their effectiveness remains unclear. Can these systems detect varying degrees of augmentation?
To address these questions, we introduce Semi-Truths, featuring 27, 600 real images, 223, 400 masks, and 1, 472, 700… See the full description on the dataset page: https://huggingface.co/datasets/semi-truths/Semi-Truths.bdnew_tar_s2d0sdoml-lite
SDOML-lite
SDOML-lite is a lightweight alternative to the SDOML dataset specifically designed for machine learning applications in solar physics, providing continuous full-disk images of the Sun with magnetic field and extreme ultraviolet data in several wavelengths. The data source is the Solar Dynamics Observatory (SDO) space telescope, a NASA mission that has been in operation since 2010.
NASA’s SDO mission has generated over 20 petabytes of high-resolution solar imagery… See the full description on the dataset page: https://huggingface.co/datasets/oxai4science/sdoml-lite.STRIDE-QA-Dataset
STRIDE-QA Dataset
📦 Dataset
STRIDE-QA is a large-scale visual question answering (VQA) dataset for physically grounded spatiotemporal reasoning in autonomous driving. Constructed from 100 hours of multi-sensor driving data in Tokyo, it offers 16 M QA pairs over 270 K frames with dense annotations including 3D bounding boxes, segmentation masks, and multi-object tracks.
Category
Description
Object-centric Spatial QA
Spatial relations between two… See the full description on the dataset page: https://huggingface.co/datasets/turing-motors/STRIDE-QA-Dataset.SceneNetRGBD
SceneNetRGBD
This is a mirror of the SceneNetRGBD dataset. The original download links at https://robotvault.bitbucket.io/scenenet-rgbd.html are no longer available, so this copy is provided here for research convenience.
Files
File
Description
SceneNetRGBD-val.tar.gz
Validation set (~15 GB)
train_0.tar.gz – train_16.tar.gz
Training set split into 17 shards (~16 GB each, ~277 GB total)
Citation
If you use this dataset, please cite the original… See the full description on the dataset page: https://huggingface.co/datasets/haotongl/SceneNetRGBD.bd_new_s2ScaleDF
Summary
This is the dataset proposed in our paper Scaling Laws for Deepfake Detection.
ScaleDF is the largest dataset in the deepfake detection domain to date. It contains over 5.8 million real images from 51 different datasets (domains) and more than 8.8 million fake images generated by 102 deepfake methods.
Using ScaleDF, we observe power-law scaling similar to that shown in large language models (LLMs). Specifically, the average detection error follows a predictable… See the full description on the dataset page: https://huggingface.co/datasets/WenhaoWang/ScaleDF.WildGUI_Screenshots
WildGUI Screenshots (part16–19)
This repository hosts the screenshot images for part16–part19 of
WildGUI, the dataset introduced by Video2GUI. It extends the main release at
xwm/WildGUI, which already
contains all annotations plus the screenshots for part1–part15.
The two repositories are split as follows:
Repository
Annotations
Screenshots
xwm/WildGUI
All parts (JSONL)
part1–part15
joker-112/WildGUI_Screenshots (this repo)
—
part16–part19
So the annotations… See the full description on the dataset page: https://huggingface.co/datasets/joker-112/WildGUI_Screenshots.font-square-v2
Accessing the font-square-v2 Dataset on Hugging Face
The font-square-v2 dataset is hosted on Hugging Face at blowing-up-groundhogs/font-square-v2. It is stored in WebDataset format, with tar files organized as follows:
tars/train/: Contains {000..499}.tar shards for the main training split.
tars/fine_tune/: Contains {000..049}.tar shards for fine-tuning.
Each tar file contains multiple samples, where each sample includes:
An RGB image (.rgb.png)
A black-and-white image (.bw.png)… See the full description on the dataset page: https://huggingface.co/datasets/blowing-up-groundhogs/font-square-v2.wds_sun397InternVL-SA1B-Caption-WebDatasetThis repo contains the recaptioned SA1B images in webdataset format. The recaptioned prompts are from https://huggingface.co/datasets/OpenGVLab/InternVL-SA-1B-Caption
nyu-depthv2-wds
Dataset Card for nyu-depthv2-wds
This is the NYU DepthV2 dataset, converted into the webdataset format. https://huggingface.co/datasets/sayakpaul/nyu_depth_v2/
There are 47584 samples in the training split, and 654 samples in the validation split.
I shuffled both the training samples, and the validation samples.
I also cropped 16 pixels from all sides of the image, and depth image. I did this because there is a white border around all images.
This is an example of the border… See the full description on the dataset page: https://huggingface.co/datasets/adams-story/nyu-depthv2-wds.SPI-2M
SPI-2M
We introduce Stylized Pathology Images SPI-2M for stain normalisation via neural style transfer in histopathology.
For full details on dataset sourcing, creation etc please see our paper
Dataset download
The data repo of this repository is organised as follows:
sources: contains the 4096 curated source images zipped together
targets: contains the 512 target images zipped together
stylized: contains 512 .npy files, each has the same index as a corresponding target… See the full description on the dataset page: https://huggingface.co/datasets/R-J/SPI-2M.MegaSynth-webdatasetfastmap_sfm
Fastmap evaluation suite.
You only need the databases to run fastmap. Download the images if you want to produce colored point cloud.
Download the subset of data you want to your local directory.
huggingface-cli download whc/fastmap_sfm --repo-type dataset --local-dir ./ --include 'databases/tnt_*' 'ground_truths/tnt_*'
or use the python interface
from huggingface_hub import hf_hub_download, snapshot_download
snapshot_download(
repo_id="whc/fastmap_sfm", repo_type='dataset'… See the full description on the dataset page: https://huggingface.co/datasets/whc/fastmap_sfm.objaverse_rendering_set