datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
everyday-manipulation-3d-raw
Everyday Manipulation 3D (raw RGB-D)
1,513 clips · 10.28 hours · 279 GiB · 4 participants · 10 manipulation tasks · 42 recording sittings
Chest-mounted iPhone Pro capture of everyday two-handed manipulation by
CaryX AI. Clips were recorded with
Record3D, an iOS app that captures the
iPhone's LiDAR RGB-D stream. Each clip is the app's .r3d recording with the
audio track removed; the sensor streams are unmodified: synchronised RGB,
metric LiDAR depth, per-frame ARKit 6-DoF camera… See the full description on the dataset page: https://huggingface.co/datasets/CaryxAI/everyday-manipulation-3d-raw.3d-models-for-isaac-sim-dataset
Dataset of 3D models for Isaac Sim (USDZ)
🇬🇧 English Description
This dataset contains a collection of 3D models converted to the .usdz format, featuring proper Semantic Labeling. These assets are optimized for generating synthetic training data using NVIDIA Isaac Sim and NVIDIA Replicator.
Primary Use Case: Training object detection and segmentation models (e.g., YOLO, RT-DETR, Mask R-CNN).
Class List
The dataset includes the following 30 semantic… See the full description on the dataset page: https://huggingface.co/datasets/barszot/3d-models-for-isaac-sim-dataset.OpenASL_3D
This is a Large-Scale 3D Datset for Continuous American Sign Language
3D-RAD
[ 🎯 NeurIPS 2025 ] 3D-RAD 🩻: A Comprehensive 3D Radiology Med-VQA Dataset with Multi-Temporal Analysis and Diverse Diagnostic Tasks
📢 News
What's New in This Update 🚀
2025.10.23: 🔥 Updated the latest version of the paper!
2025.09.19: 🔥 Paper accepted to NeurIPS 2025! 🎯
2025.05.16: 🔥 Set up the repository and committed the dataset!
🔍 Overview
💡 In this repository, we present the dataset for "3D-RAD: A… See the full description on the dataset page: https://huggingface.co/datasets/Tang-xiaoxiao/3D-RAD.3DTime
3DTime dataset: (sample of) A Large Dataset of Multivariate Time-Series for 3D-printing Duration
This dataset is a small sample of the 3DTime dataset, for which the paper is currently under review for NeurIPS Datasets and Benchmarks 2026.
This smaller version contains:
82 3D models (~0.08% of the full dataset), their corresponding sliced G-code, compressed annotated G-code, and binary vectorized files
A total of 5,855,369 G-code instructions (~0.07% of the full dataset)… See the full description on the dataset page: https://huggingface.co/datasets/3DTimeDataset/3DTime.3D-DefectBench
3D-DefectBench
A controlled benchmark for evaluating vision-language models (VLMs) as fine-grained judges of
defects in text-to-3D generation.
3D-DefectBench is a VLM-as-a-judge benchmark for detecting fine-grained defects in textured 3D
meshes. It lets you measure how well any VLM judge aligns with human judgment: run your judge over the
assets and score its predictions against the human defect labels provided here.
Each example pairs a text prompt with a generated, textured 3D… See the full description on the dataset page: https://huggingface.co/datasets/zzhao0500/3D-DefectBench.SceneFly
CaR
Compression and Retrieval: Implicit Memory Retrieval for Video World Models
Zhan Peng1,2,
Jie Ma2,
Huiqiang Sun1,
Chong Gao2,3,
Zhijie Xue1,
Zhiyu Pan1,
Zhiguo Cao1*,
Jun Liang2*,
Jing Li2
1Huazhong University of Science and Technology
2HUJING Digital Media & Entertainment Group
3Sun Yat-sen University
*Corresponding author
SceneFly
SceneFly is a curated video dataset organized by synthetic 3D scenes. Each selected video… See the full description on the dataset page: https://huggingface.co/datasets/Orange-3DV-Team/SceneFly.3DA-VTG
3DA-VTG
3DA-VTG is a visuo-tactile grasp-stability dataset prepared for the public
SGA-GSN release. It contains paired visual and tactile observations, object-level
metadata, and binary grasp-stability labels.
License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0
International (CC BY-NC-SA 4.0). The dataset is derived from GraspNet-1Billion
assets and follows the GraspNet non-commercial redistribution terms. Commercial
use requires permission from the GraspNet team.… See the full description on the dataset page: https://huggingface.co/datasets/robotic-vt-grasp-project/3DA-VTG.3DRAG-Bench
3DRAG-Bench
This dataset contains 100 curated 3D object assets for 3DRAG 3D editing experiments.
Each object is stored as a GLB mesh together with a cleaned editing specification.
Dataset Structure
.
+-- README.md
+-- LICENSE
+-- .gitattributes
+-- metadata.csv
+-- name_mapping.csv
`-- assets/
`-- <asset_name>/
+-- model.glb
`-- dataset_input_clean.json
Files
assets/<asset_name>/model.glb: GLB asset file.… See the full description on the dataset page: https://huggingface.co/datasets/AeTherRaIn/3DRAG-Bench.3DSRBench
3DSRBench Circular Evaluation Package
This upload contains the processed TSV required by PhysBrainEvalKit for 3DSRBench circular evaluation. The TSV embeds the evaluation images as base64 data, so no separate image archive is required.
Files in this repository
3dsrbench_v1_vlmevalkit_circular.tsv: processed evaluation data used by PhysBrainEvalKit.
compute_3drbench_results_circular.py: optional standalone result computation script.
.gitattributes: large-file… See the full description on the dataset page: https://huggingface.co/datasets/VLyb/3DSRBench.3D-CT-report-generationwelsh-speech-3d-meshes
Welsh Speech Dataset - 3D Facial Meshes
3D facial reconstructions from the Welsh Speech Dataset.
Contents
3D meshes (.obj files) - One per frame
Texture maps (.png files) - Fused left-right stereo images from 3DMD
Captured using 3DMD 6-camera system
~330 zip files (one per speaker-phrase sequence)
File Structure
Files are organized as zip archives in the meshes/ directory, one zip per speaker-phrase sequence:
meshes/
├── speaker_01_phrase_01.zip
├──… See the full description on the dataset page: https://huggingface.co/datasets/arvinsingh/welsh-speech-3d-meshes.3DOpenASL
This is a Large-Scale 3D Datset for Continuous American Sign Language
3DKoreanMelon
Korean Melon 3D Growth Sequences
Per-fruit 3D observation sequences of Korean melon (Cucumis melo L. var. makuwa) grown
on the plant, each paired with a post-harvest scan of the same fruit and with vernier
caliper measurements taken at every visit.
Fruits were revisited every two to three days over a full growing period and imaged in
place, so each sequence follows one identified fruit as it enlarges while foliage occludes
a different part of it at each visit.… See the full description on the dataset page: https://huggingface.co/datasets/Sungjay/3DKoreanMelon.3D-PAQA
3D-PAQA — Preference-Aligned 3D Quality Assessment
Preference-aligned perceptual quality labels for 240,636 Objaverse assets, rated on
six perceptual criteria. The goal of 3D-PAQA is to move beyond synthetic-distortion
3D-QA benchmarks and provide human-preference-aligned quality scores for real,
human-created 3D assets, at a scale usable for training and benchmarking automatic
quality evaluators. Drawn from a 264,966-asset Objaverse corpus.
train.csv — 216,540 labeled assets… See the full description on the dataset page: https://huggingface.co/datasets/JiHyuk-Byun/3D-PAQA.3D-Printable-Guitar-Modelsindoor-smartphone-3d-reconstruction-control
Indoor Smartphone 3D Reconstruction Control Dataset
Набор данных подготовлен для сравнения методов восстановления 3D-сцены в помещении по короткому видео со смартфона. Он содержит три небольшие indoor-сцены, очищенные публикационные видеоролики, подвыборки кадров, ручную CVAT-разметку и физически измеренные контрольные расстояния.
Датасет предназначен для оценки геометрической согласованности результатов 3D-реконструкции. Разметка не является обучающей dense-разметкой глубины:… See the full description on the dataset page: https://huggingface.co/datasets/Maksonchek/indoor-smartphone-3d-reconstruction-control.colon-3d-crc13D_volumes_EMPIAR106483D volumes of the EMPIAR-10648 validation dataset generated using the original images, low-res images, and CryoGEN reconstructed images.
cadquery-creating-basic-2d-and-3d-formsfirm-shoe-3dd73b
firm-shoe-3dd73b
Synthetic weather test data: 40 rows in data.csv.
All values are randomly generated fictional examples, not real observations, products, or user activity. Intended only for CSV loading and pipeline tests; not suitable for scientific or business conclusions. Columns are sampled independently and do not model real-world correlations.
Fields
sample_id: random identifier for this generated sample.
row_id: sequential row number starting at 1.… See the full description on the dataset page: https://huggingface.co/datasets/mark-miller/firm-shoe-3dd73b.superconductor-3dsc
3DSC Superconductor Dataset
This dataset contains 5,773 superconducting materials for training ML models to predict critical temperatures.
Dataset Summary
Total Materials: 5,773
Tc Range: 0-132 K
Splits: Train (4,041) / Val (866) / Test (866)
Data Fields
material_id: Unique identifier
cif_id: Materials Project ID
Tc: Critical temperature (K)
formula: Chemical formula
cif_path: Path to CIF file
Usage
from datasets import load_dataset
dataset =… See the full description on the dataset page: https://huggingface.co/datasets/shreyaspullehf/superconductor-3dsc.3dct_lung_vqa_with_reasoningMedUI_3DSlicer_CSV
3D Slicer Medical Imaging GUI Benchmark Dataset (CSV Format)
Dataset Description
This dataset contains 315 end-to-end GUI automation tasks for 3D Slicer medical imaging software, focusing on MRI brain analysis workflows.
Dataset Summary
Total Tasks: 315
Total Images: 100 unique screenshots (file paths only)
Application: 3D Slicer (medical imaging software)
Domain: Medical imaging, MRI brain analysis
Format: CSV with file paths (ultra memory-efficient)… See the full description on the dataset page: https://huggingface.co/datasets/rishuKumar404/MedUI_3DSlicer_CSV.3D_Synthetic_Petroleum_Derived_GEMS
3D Synthetic Petroleum-Derived GEMS (MVP Release)
📌 Dataset Overview
This dataset contains 197 elite, highly complex 3D molecular structures derived from petroleum fractions. Designed specifically for petrochemicals, materials science, organic semiconductors, and specialized additives, these compounds represent a curated "Golden Fund" of stable, complex hydrocarbons.
Unlike drug-like molecules, this dataset focuses on polycyclic architectures, rigid 3-ring… See the full description on the dataset page: https://huggingface.co/datasets/nadizik/3D_Synthetic_Petroleum_Derived_GEMS.3d_imagecrowdsourced_3dgs_cleaned3D-DST-captions
3D-DST-captions
As part of our data release in 3D-DST, we present MiniGPT4-generated captions for all 1000 classes in ImageNet-1k.
See wufeim/DST3D for synthetic data generation with 3D annotations using the captions here.
These captions can be used to produce other synthetic datasets for fair comparisons between different data generation procedures.
3dct_better_vqa3D_Continuous_ASL
This is a Large-Scale 3D Datset for Continuous American Sign Language