datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
SpatialForge
SpatialForge-10M
SpatialForge: Bootstrapping 3D-Aware Spatial Reasoning from Open-World 2D Images
📑 Paper
Zishan Liu, Ruoxi Zang, Yanglin Zhang, Wei Liu, Yin Zhang, Jian Yao, Jiayin Zheng, Zhengzhe Liu
Lingnan University · XPENG Robotics
📦 SpatialForge-10M
A large-scale vision-language dataset designed for 3D-aware spatial perception and reasoning from open-world 2D images.
SpatialForge-10M contains over 10 million QA pairs generated from 2.8 million curated… See the full description on the dataset page: https://huggingface.co/datasets/shana643/SpatialForge.SpatialReward-Train
SpatialReward-Train Dataset
Training data for SpatialReward, containing two stages:
Split
Description
Annotation file
Images
rl/
RL training data
data.json
images.tar
sft/
SFT training data (~260k)
data.jsonl
images_part_aa ~ images_part_am
Download & Extract
RL Data
# Download
huggingface-cli download SpatialReward/SpatialReward-Train \
rl/data.json rl/images.tar --repo-type dataset
# Extract images
cd rl/
tar -xf images.tar… See the full description on the dataset page: https://huggingface.co/datasets/SpatialReward/SpatialReward-Train.map-spatial-benchmark
Map-based Spatial Reasoning Benchmark
A multi-view map-based spatial reasoning benchmark. Each row is one multiple-choice
question instance over a registered map image; models must answer with a single option
letter. Four tasks (T1–T4), four base-map views, and controlled evidence conditions
(direct / query / oracle) and world perturbations (transform / world layers) allow
fine-grained analysis of spatial reasoning robustness.
Task overview
Task
Question… See the full description on the dataset page: https://huggingface.co/datasets/mapspatial/map-spatial-benchmark.embodied-spatial-reasoning
Embodied Spatial Reasoning Tasks
Dataset Description
This dataset is part of the embodied-spatial-reasoning project, where the agent has to actively explore the environment to determine if certain spatial relationships hold true. The tasks involve spatial reasoning with various objects and scenes. Each task includes a query about the spatial relationships between objects within a scene, which the agent must verify through exploration.
Dataset Structure
The… See the full description on the dataset page: https://huggingface.co/datasets/thanhqt2002/embodied-spatial-reasoning.spatialtestSpatialCLI-Data
SpatialCLI-Bench
Viewer data: data/eval/SpatialCLI-Bench/full.jsonl
Split: test
Number of records: 516
Shared visual assets: data/assets/
Paths embedded in the records are relative to the repository root.
Paper
This dataset accompanies the SpatialCLI paper.
Code
The code and model checkpoints are available at SpatialCLI GitHub repository.
SpatialMQAWelcome to explore our work titled "Can Multimodal Large Language Models Understand Spatial Relations".arXiv link: https://arxiv.org/abs/2505.19015.For more information about the paper and the SpatialMQA dataset, please visit our GitHub repository at https://github.com/ziyan-xiaoyu/SpatialMQA.
license: cc-by-4.0
SpatialGen-Testset
SpatialGen Testset
This repository contains the test set for SPATIALGEN: Layout-guided 3D Indoor Scene Generation, a novel multi-view multi-modal diffusion model for generating realistic and semantically consistent 3D indoor scenes.
Project page | Paper | Code
We provide a test set of 48 preprocessed point clouds and their corresponding GT layouts, multi-view images are cropped from the high-resolution panoramic images.
Folder Structure
Outlines of the dataset files:… See the full description on the dataset page: https://huggingface.co/datasets/manycore-research/SpatialGen-Testset.grounded-visual-spatial-reasoning
Grounded Visual Spatial Reasoning
Code for generating the annotations can be found here: github.com
Dataset Summary
This dataset extends the Visual Spatial Reasoning (VSR) dataset with visual grounding annotations: each caption is annotated with COCO-category object mentions, their positions , and corresponding bounding boxes in the image.
Data instance
Each sample instance has the following structure:
Field
Type
Description
image_file
string… See the full description on the dataset page: https://huggingface.co/datasets/tomhodemon/grounded-visual-spatial-reasoning.SpatialGen-Bench
Benchmark
Each record retains its source-task metric target as integer, text, point, mask, or polyline. The frozen visual-answer contract is available at protocols/spatialgen_bench.yaml, with runtime parsers in ProVisE.
Quick Start
from datasets import load_dataset
dataset = load_dataset("wx91726/SpatialGen-Bench", split="test")
print(dataset[0])
Download the complete media and evaluation package for local evaluation:
hf download… See the full description on the dataset page: https://huggingface.co/datasets/wx91726/SpatialGen-Bench.SpatialLadder-26k
SpatialLadder-26k
This repository contains the SpatialLadder-26k, introduced in SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models.
Dataset Description
SpatialLadder-26k is a large-scale training dataset designed to develop spatial perception and reasoning capabilities in Vision-Language Models (VLMs). It contains 26,610 multimodal samples spanning four complementary task categories, forming a… See the full description on the dataset page: https://huggingface.co/datasets/hongxingli/SpatialLadder-26k.Spatial-SSRL-81k
Spatial-SSRL-81k
📖Paper| 🏠Github |🤗Spatial-SSRL-7B Model |
🤗Spatial-SSRL-3B Model | 🤗Spatial-SSRL-Qwen3VL-4B Model |
🤗Spatial-SSRL-81k Dataset | 📰Daily Paper
Spatial-SSRL-81k is a training dataset for enhancing spatial understanding in large vision-language models. It contains 81,053 samples of five pretext tasks for self-supervised learning, offering simple, intrinsic supervision that scales RLVR efficiently.
📢 News
🚀 [2026/04/05] We have released… See the full description on the dataset page: https://huggingface.co/datasets/cvis-tmu/Spatial-SSRL-81k.salabs-robotics-spatial-topology-v9
🤖 SALabs 768-D Continuous Lie SE(3) Robotics & Spatial Manifold Topology Dataset (v9.0)
[!IMPORTANT]
💳 Click Here to Purchase Enterprise Commercial License ($1,500 USD) & Instant 391.56MB Master DownloadInstant download of the full 391.56MB Enterprise JSONL matrix containing 50,000+ continuous Lie $SE(3)$ manifold trajectories, singularity-free Bishop Frame metrics, and commercial license certificate.
🌟 Executive Summary
The SALabs Robotics & Spatial… See the full description on the dataset page: https://huggingface.co/datasets/suitai/salabs-robotics-spatial-topology-v9.Spatial-Awareness-Datasetstructured3d-spatiallm
Structured3D-SpatialLM Dataset
Structured3D dataset preprocessed in SpatialLM format for layout estimation with LLMs.
Overview
This dataset is derived from Structured3D 3,500 synthetic house designs created by professional designers, preprocessed and formatted specifically for SpatialLM training.
Point clouds and layouts are derived from the RoomFormer data preprocessing script.
Data Extraction
Point clouds and layouts are compressed in zip files. To… See the full description on the dataset page: https://huggingface.co/datasets/ysmao/structured3d-spatiallm.salabs-virtual-spatial-digitaltwin-v8
🌐 SALabs 10,000,000-Node 3D Virtual Spatial & Digital Twin Avatar Kinematics Dataset (v8.0)
[!IMPORTANT]
💳 Click Here to Purchase Enterprise Commercial License ($2,000 USD) & Instant 8.0GB Master DownloadInstant download of the complete 8.0GB master archive containing 10,000,000 verified 3D spatial nodes, 18-DoF avatar kinematics, B-spline 4D motion tensors, Laplace-Beltrami spectral resonance, and commercial license certificate.
🌟 Executive Summary
The… See the full description on the dataset page: https://huggingface.co/datasets/suitai/salabs-virtual-spatial-digitaltwin-v8.arkitscenes-spatiallm
ARKitScenes-SpatialLM Dataset
ARkitScenes dataset preprocessed in SpatialLM format for oriented object bouding boxes detection with LLMs.
Overview
This dataset is derived from ARKitScenes 5,047 real-world indoor scenes captured using Apple's ARKit framework, preprocessed and formatted specifically for SpatialLM training.
Data Extraction
Point clouds and layouts are compressed in zip files. To extract the files, run the following script:
cd arkitscenes-spatiallm… See the full description on the dataset page: https://huggingface.co/datasets/ysmao/arkitscenes-spatiallm.CT-SpatialVQA
CT-SpatialVQA
CT-SpatialVQA is a benchmark for evaluating semantic-spatial reasoning in 3D CT, derived from the CT-RATE dataset.
Stat
Value
CT volumes
1,601 (CT-RATE validation split)
QA pairs
9,077
Spatial Categories
Questions are designed to require explicit spatial grounding across six categories:
Laterality & Bilateral Symmetry — left vs. right, bilateral involvement
Longitudinal (Vertical) Position — superior/inferior, upper/middle/lower… See the full description on the dataset page: https://huggingface.co/datasets/Mashrafi2827/CT-SpatialVQA.spatial-training-full-release-20260604
Spatial Training Full Data Release
Full data staging directory for our current Cambrian-P / SSR-style Spatial VLM reproduction work.
The directory contains the lightweight reproduction pack plus raw compressed training archives. Local staging uses hardlinks where possible, but upload payload is the full dataset.
Size
Logical payload: 1076.281 GiB
Files: 254
Max single file: 18.0 GiB
VSI-590K raw payload: 216.777 GiB
Cambrian-S-3M raw payload: 858.004 GiB… See the full description on the dataset page: https://huggingface.co/datasets/Gradygu3u/spatial-training-full-release-20260604.Spatial-Scene-Synthetic-Datasetspatial_mosaic_vqa
SpatialMosaic: A Multi-View VLM Dataset for Partial Visibility
Description
SpatialMosaic is a multi-view visual question answering dataset for evaluating spatial reasoning under partial visibility, occlusion, and low-overlap views. It pairs indoor ScanNet++ and outdoor Waymo scene references with multi-frame VQA annotations. Questions require models to combine fragmented evidence across 2-5 views, rather than answering from a single image. The tasks… See the full description on the dataset page: https://huggingface.co/datasets/jmkey/spatial_mosaic_vqa.pitvqa-comprehensive-spatial
PitVQA Comprehensive Spatial Dataset
High-fidelity surgical spatial localization dataset for training vision-language models on pituitary surgery instrument and anatomy detection.
🔗 GitHub: https://github.com/matheus-rech/pit_project
🤖 Trained Model: mmrech/pitvqa-qwen2vl-spatial
📄 Original Dataset: UCL Research Data Repository
Dataset Description
This dataset contains 10,139 surgical frames with precise spatial annotations for instrument localization and anatomy… See the full description on the dataset page: https://huggingface.co/datasets/mmrech/pitvqa-comprehensive-spatial.SpatialEvo-160K
SpatialEvo: Self-Evolving Spatial Intelligence via Deterministic Geometric Environments
SpatialEvo-160K
Dataset Description
SpatialEvo-160K is an offline spatial reasoning QA dataset generated by the Deterministic Geometric Environment (DGE) from SpatialEvo: Self-Evolving Spatial Intelligence via Deterministic Geometric Environments. This dataset is not used in the SpatialEvo training pipeline reported in the paper; it is released… See the full description on the dataset page: https://huggingface.co/datasets/lidingm/SpatialEvo-160K.SpatialGen-Testset
SpatialGen Testset
This repository contains the test set for SPATIALGEN: Layout-guided 3D Indoor Scene Generation, a novel multi-view multi-modal diffusion model for generating realistic and semantically consistent 3D indoor scenes.
Project page | Paper | Code
We provide a test set of 48 preprocessed point clouds and their corresponding GT layouts, multi-view images are cropped from the high-resolution panoramic images.
Folder Structure
Outlines of the dataset… See the full description on the dataset page: https://huggingface.co/datasets/BenjaminChai579/SpatialGen-Testset.SpatialStackDataspatialvlm_routeplanlibero_plus_spatial
libero_plus_spatial: detailed LeRobot v3.0
This dataset was converted from the LIBERO Plus LeRobot v2.1 libero_plus_spatial partition.
The original 8D state and 7D action vectors are preserved exactly as
raw_state.ref_state and raw_action.ref_action. Canonical low-dimensional fields follow
failure_rollout_data/dataset.md; debug.gripper_eef_* contains the ground-truth next-step
relative EEF motion for inspection.
Required camera transform for canonical training
The… See the full description on the dataset page: https://huggingface.co/datasets/typoverflow/libero_plus_spatial.spatial_reasoningspatial-relationsSpatial-Scene-Dataset-Real