datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
3DCode
Project page
Paper
Code
3dcodebench.com
arXiv:2606.01057
gaoypeng/3dcodebench
News
[06/01/2026] Paper released on arXiv: 3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code.
Note. This is an open-source reproduction of 3DCodeBench.
⚠️ Under final check. The 3DCodeData/ code is still undergoing final
quality review and may contain occasional issues (non-executable scripts, mismatched
captions/renders, or imperfect geometry). If you run… See the full description on the dataset page: https://huggingface.co/datasets/YipengGao/3DCode.3d-front-ar3DSpatialBench
3DSpatialBench
This dataset repository contains a processed CSV file for 3D spatial benchmarking.
File: filtered_processed_.csv
Source path (local): /pfs/gaohongcheng/3ddata/filtered_processed_.csv
Please update this README with schema, column descriptions, and licensing info.
3DSRBench
3DSRBench: A Comprehensive 3D Spatial Reasoning Benchmark
We present 3DSRBench, a new 3D spatial reasoning benchmark that significantly advances the evaluation of 3D spatial reasoning capabilities of LMMs by manually annotating 2,100 VQAs on MS-COCO images and 672 on multi-view synthetic images rendered from HSSD. Experimental results on different splits of our 3DSRBench provide valuable findings and insights that will benefit future research on 3D spatially… See the full description on the dataset page: https://huggingface.co/datasets/ccvl/3DSRBench.ReVSI
ICML 2026
Yiming Zhang1*,
Jiacheng Chen1*,
Jiaqi Tan1,
Yongsen Mao2,
Wenhu Chen3,
Angel X. Chang1,4
1 Simon Fraser University
2 Hong Kong University of Science and Technology
3 University of Waterloo
4 Alberta Machine Intelligence Institute (Amii)
This repository contains the ReVSI benchmark and dataset, introduced in ReVSI: Rebuilding Visual Spatial Intelligence Evaluation for… See the full description on the dataset page: https://huggingface.co/datasets/3dlg-hcvc/ReVSI.SSR-3DFRONT
SSR-3DFRONT: Structured Scene Representation for 3D Indoor Scenes
This dataset provides a processed version of the 3D-FRONT dataset with structured scene representations for text-driven 3D indoor scene synthesis and editing.
Mor information about ReSpace: http://respace.mnbucher.com
For detailed usage instructions, training details, and examples, see the associated repository: https://github.com/GradientSpaces/respace
Our model weights for SG-LLM:… See the full description on the dataset page: https://huggingface.co/datasets/gradient-spaces/SSR-3DFRONT.ramanv-image-real-3d-rendersheat_conduction_3d_v0This dataset is part of the EngiBench toolkit: https://github.com/IDEALLab/EngiBench.
3DSRBench_lmmseval
3DSRBench (lmms-eval compatible)
This is a reformatted version of 3DSRBench for compatibility with lmms-eval.
Dataset Description
3DSRBench is a comprehensive 3D spatial reasoning benchmark that evaluates the 3D spatial reasoning capabilities of Large Multimodal Models (LMMs). It includes 2,100 VQAs on MS-COCO images and 672 on multi-view synthetic images rendered from HSSD.
Subsets
This dataset contains two subsets:
1. 3dsr… See the full description on the dataset page: https://huggingface.co/datasets/oscarqjh/3DSRBench_lmmseval.STS-3D-Tooth
STS-3D-Tooth
The 3D Cone-Beam CT (CBCT) subset of the STS (Semi-supervised Teeth
Segmentation) multi-modal dental dataset, as released in
Wang et al., Scientific Data 12, 117 (2025)
and used in the MICCAI 2023/2024 STS Challenges.
The companion 2D panoramic X-ray subset is hosted at
Angelou0516/STS-2D-Tooth.
Dataset Summary
Field
Details
Modality
Cone-Beam CT (CBCT), NIfTI (.nii.gz)
Body Part
Teeth (32 permanent teeth, FDI numbering)
Volumes
371… See the full description on the dataset page: https://huggingface.co/datasets/MedOtter/STS-3D-Tooth.3DSRBench_synthetic3d-front-ar-packed
PixARMesh Training Dataset
Project Page | Paper | GitHub
This repository contains the training dataset for PixARMesh, a method to autoregressively reconstruct complete 3D indoor scene meshes directly from a single RGB image. Unlike prior methods that rely on implicit signed distance fields, PixARMesh jointly predicts object layout and geometry within a unified model, producing coherent and artist-ready meshes in a single forward pass.
Dataset Preparation
According… See the full description on the dataset page: https://huggingface.co/datasets/zx1239856/3d-front-ar-packed.delete_meFLOW-3D-Single-Track
FLOW-3D-Single-Track
This dataset consists of LPBF (Laser Powder Bed Fusion) melt pool simulations performed using FLOW-3D. Each row is one timestep from one simulation case, containing process parameters, mesh metadata, a per-timestep field volume, and image previews.
Dataset Configs
Load a specific field config:
from datasets import load_dataset
ds = load_dataset("ppak10/FLOW-3D-Single-Track", "temperature")
Available configs:
Config
Volume Shape… See the full description on the dataset page: https://huggingface.co/datasets/ppak10/FLOW-3D-Single-Track.Face-Depth-3D
Face-Depth-3D
Face-Depth-3D is a high-quality dataset designed for face depth estimation and 3D face reconstruction. The dataset is a combination of both male and female face portraits, providing a diverse collection of facial appearances for training and evaluating modern computer vision and image-to-3D models. Each sample contains an RGB face image, a dense facial depth map, and a corresponding 3D mesh in GLB format, allowing direct supervision for… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/Face-Depth-3D.CV-Bench-3D-gt3dloc-v3audio-3dvgthermoelastic_3d_v0eval_pi0_3D_infe_white_bowlThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 15,
"total_frames": 8315,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:15"
},
"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/tersooawai/eval_pi0_3D_infe_white_bowl.COCO_3D
COCO 3D 扩展数据集
数据集描述
本数据集基于 COCO 2017 训练集，添加了深度和方向信息：
深度：通过 depth-pro 模型预测的绝对深度（米）。
方向：通过 orient-anything 模型预测的方位角（azimuth/polar）。
原始标注：包括类别、2D 框、RLE 分割掩码。
使用示例
from datasets import load_dataset
dataset = load_dataset("lidaiqiang/COCO_3D", split="train")
sample = dataset[0]
print(sample["objects"][0]["depth"]) # 打印第一个物体的深度
字段说明
字段名
类型
描述
image
Image
PIL 格式的原始图像
objects
Sequence
物体标注列表（见下表）
→ class
str
物体类别（如 "cat"）
→… See the full description on the dataset page: https://huggingface.co/datasets/lidaiqiang/COCO_3D.Female-Face-Depth-3D
Female-Face-Depth-3D
Female-Face-Depth-3D is a high-quality dataset designed for female face depth estimation and 3D face reconstruction. The dataset contains paired RGB face images, dense facial depth maps, and corresponding 3D meshes in GLB format, making it suitable for training and evaluating modern computer vision and image-to-3D models. Every sample provides a direct correspondence between a facial photograph, its reconstructed depth representation, and an associated 3D… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/Female-Face-Depth-3D.oasis-distillation-dataset-3d3D-PC
Summary
Visual perspective taking (VPT), the ability to accurately perceive and reason about the perspectives of others, is an essential feature of human intelligence.
Deep neural networks (DNNs) may be a good candidate for modeling VPT and its computational demands in light of a growing number of reports indicating that DNNs gain the ability to analyze 3D scenes after training on large static-image datasets.
We developed the 3D perception challenge (3D-PC) for comparing 3D… See the full description on the dataset page: https://huggingface.co/datasets/3D-PC/3D-PC.hitek-3digit-shardstest_multiview_3d_reconstructionThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "so100_follower",
"total_episodes": 1,
"total_frames": 757,
"total_tasks": 1,
"total_videos": 2,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:1"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/jccj/test_multiview_3d_reconstruction.agentic-3d-worlds
Agentic 3D Worlds
Agent traces of 3D objects modeled primarily in Blender with bpy, paired with their creation code, .blend scenes, self-contained GLB exports and rendered previews.
Share 3D World Agent Run is a skill that collects, redacts PII, and validates these files, then submits a PR after user confirmation. No contributor terminal commands or dataset setup are needed. Please use it to make PRs to this dataset to contribute!
Runs use short descriptive titles and omit… See the full description on the dataset page: https://huggingface.co/datasets/suvadityamuk/agentic-3d-worlds.seu-3dgs
Single-Event Upsets in 3D Gaussian Splatting Rendering
Artifacts for the paper Single-Event Upsets in 3D Gaussian Splatting Rendering: Bit-Level Criticality, Spatial Extent, and a Parallel Support Guard (F. Alpay and B. Basaran).
A trained 3DGS model is a large floating-point array resident in GPU memory, so a single-event upset is one flipped bit in one parameter. This repository releases the fault-injection engine, the trained models, the per-cell aggregated records of more… See the full description on the dataset page: https://huggingface.co/datasets/Lightcap/seu-3dgs.Face-3D-Unified-Preferences
Face-3D-Unified-Preferences
Face-3D-Unified-Preferences is a high-quality dataset designed for human face depth estimation, 3D face reconstruction, and unified preference learning. The dataset is a mixture of male and female human face portraits, providing a diverse collection of facial appearances, identities, poses, and expressions for training modern computer vision and multimodal AI models. Each sample contains an RGB face image, a dense facial depth map, and a… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/Face-3D-Unified-Preferences.test_multiview_3d_reconstruction_3camsThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "so100_follower",
"total_episodes": 1,
"total_frames": 375,
"total_tasks": 1,
"total_videos": 3,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:1"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/jccj/test_multiview_3d_reconstruction_3cams.hitek-3digit-shards