datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Motion-o-MCoT-PLM-motion-keyframes
Motion-o-MCoT (PLM + motion keyframes)
Subset of STGR: STR_plm_rdcap rows with <motion in reasoning_process, plus sharded keyframes under videos/stgr/plm/kfs/.
Train split: 3,168 examples (see export_manifest.json in the repo for exact export stats).
Keyframes: JPEGs are stored under shard subfolders (e.g. videos/stgr/plm/kfs/plm_0150/…) so each directory stays under Hugging Face file-count limits. Each key_frames[].path in the JSON is relative to videos/stgr/plm/kfs/ (e.g.… See the full description on the dataset page: https://huggingface.co/datasets/bishoygaloaa/Motion-o-MCoT-PLM-motion-keyframes.motionatlas-data
MotionAtlas-Data
MotionAtlas-Data is a large-scale dataset for region-aware motion captioning. Instead of describing a whole clip globally, each sample pairs a video with a spatiotemporal region and a precise description of the motion inside that region, reducing visual clutter and motion entanglement.
159K high-quality region-level motion captioning samples
Built with a scalable pipeline using self-bootstrap refinement to suppress fine-grained hallucinations
Designed to… See the full description on the dataset page: https://huggingface.co/datasets/maxLWSv2/motionatlas-data.Motion-Xplusplus
Data for Motion-X++: A Large-Scale Multimodal 3D Whole-body Human Motion Dataset
Here, we release our dataset, "Motion-X++: A Large-Scale Multimodal 3D Whole-body Human Motion Dataset," which includes various motion modalities. It features 2D keypoints for mesh recovery and motion generation. Additionally, we provide SMPL-X annotations that differentiate between translations and orientations in camera and world coordinate systems. The dataset also includes action descriptions… See the full description on the dataset page: https://huggingface.co/datasets/YuhongZhang/Motion-Xplusplus.Kimodo-Motion-Gen-Benchmark
Kimodo Human Motion Generation Benchmark
Kimodo Codebase, Benchmark Documentation
Dataset Description:
This dataset provides the necessary metadata to construct the suite of test cases that make up the Kimodo human motion generation benchmark. This includes test cases that evaluate text-following for the text-to-motion task, along with constraint-following for constraint-conditioned motion generation.
The benchmark is constructed from the SOMA uniform version of the… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Kimodo-Motion-Gen-Benchmark.motion-smd-data
Motion-SMD Data
Data release for "Encoder-Free Human Motion Understanding via Structured Motion Descriptions".
🌐 Project page: https://yaozhang182.github.io/motion-smd/
💻 Code: https://github.com/yaozhang182/motion-smd
🤗 LoRA adapters: https://huggingface.co/zyyy12138/motion-smd-lora
📄 Paper (arXiv): https://arxiv.org/abs/2604.21668
What's here
Four subdirectories, each with its own README.md describing files, provenance, and license:
Subdir
Contents
Our… See the full description on the dataset page: https://huggingface.co/datasets/zyyy12138/motion-smd-data.MotionBlind
MotionBlind
A contrastive benchmark for physical-motion perception in Video-LLMs.
📄 Paper: MotionBlind: Probing the Illusion of Motion Understanding in Video-LLMs
A Video-LLM can watch two clips of the same person in the same room and name every
object in both, yet fail to tell you which one moves faster, which way a hand travels,
or how far a box slides. MotionBlind is a contrastive, minimal-pairs benchmark
built to expose exactly that gap: pairs of near-identical clips that… See the full description on the dataset page: https://huggingface.co/datasets/augmentedcognitionlab/MotionBlind.minecraft-motion-action-datasetMotionEdit-TrainStarCraft-MotionMotionFix
MotionFix MotionHub Format
This dataset contains the processed MotionFix motion-editing data in the MotionHub format.
It stores paired source and target motions as SMPL-H 52-joint parameter files, plus editing instructions and split annotations.
Please also follow the license and terms of the original MotionFix dataset.
Structure
MotionFix/
├── smplh_52/
│ ├── train/
│ │ ├── 000000_000499/
│ │ ├── 000500_000999/
│ │ └── ...
│ ├── val/
│ └── test/… See the full description on the dataset page: https://huggingface.co/datasets/ZeyuLing/MotionFix.MotionMillion
🔑 Key Features
Over 2000 hours of high-quality human motion captured from web-scale human video data, covering:
Martial Arts (23.7%)
Fitness (26.4%)
Performance (17.5%)
Dance (14.9%)
Non-Human (2.9%)
Sports (2.4%)
Over 20 detailed annotations per motion, including:
Age
Body Characteristics
Movement Styles
Emotions
Environments
👨🏫 Get Started
Download the Dataset
To download the full dataset, use the following code. If you encounter any… See the full description on the dataset page: https://huggingface.co/datasets/InternRobotics/MotionMillion.comma-training-dataset-v1.0-chunksMotion324SIS-Motion-54K
✨SIS-Motion-54K✨
SIS-Motion-54K is a motion-aware instruction-tuning dataset built from the AirScape training split. It is designed to fine-tune the SIS-Motion model for joint understanding of space (environment) and self (agent motion) in embodied UAV scenarios.
Important: This dataset is strictly separated from the SIS-Bench evaluation benchmark. It contains only perception and memory tasks — no reasoning-level data — so any generalization gains on… See the full description on the dataset page: https://huggingface.co/datasets/choucsan/SIS-Motion-54K.motion_examplesminecraft-motion-coa-datasetenglish-debate-motions-utdsEnglish Debate Motions gathered by University of Tokyo Debate Society
@misc{english-debate-motions-utds,
title={english-debate-motions-utds},
author={members of the University of Tokyo Debate Society},
year={2022},
}
motion_gen_dataMotion_Dataset
Apple Arts Studios Motion Capture Dataset
Dataset Overview
The Apple Arts Studios Motion Capture Dataset is a professionally captured, full-body human-motion dataset designed for:
Artificial intelligence
Humanoid robotics
Motion generation
Animation
Simulation
Action recognition
Human-motion research
This release contains 5 hours of originally captured motion data.
Left–right mirrored versions of the original motions are also included, increasing the total… See the full description on the dataset page: https://huggingface.co/datasets/Appleartsstudios/Motion_Dataset.fineweb-ultra-mini
Dataset Card for Fineweb Ultra Mini
Fineweb Ultra Mini is a dataset derived from the original Fineweb dataset made by huggingface (see here: https://huggingface.co/datasets/HuggingFaceFW/fineweb).
The dataset focuses on extracting high quality data from the Fineweb dataset, from the 2-3% range. If you would like even more high-quality data, keep out for our next release, fineweb ultra mini pro, which focuses on the 0-1% of high quality data originally found in fineweb.… See the full description on the dataset page: https://huggingface.co/datasets/motionlabs/fineweb-ultra-mini.motiontrans
MotionTrans Dataset
[MotionTrans Main Repository]
Download
Please download this repository and unzip motiontrans_dataset.zip.
Introduction
MotionTrans is the first framework to achieve explicit end-to-end human-to-robot motion transfer, enabling motion-level policy learning directly from human data. This repository contains the MotionTrans Dataset used in our paper, including zero-shot human-robot cotraining data and finetuning data. The dataset provides 15… See the full description on the dataset page: https://huggingface.co/datasets/michaelyuanqwq/motiontrans.Full_Length_Motion_Capture_Dataset
Apple Arts Studios Full-Length Motion Capture Dataset
Dataset Overview
The Apple Arts Studios Full-Length Motion Capture Dataset is a professionally captured, full-body human-motion dataset containing 199 hours and 30 minutes of continuous motion capture data.
Unlike segmented motion datasets, this repository preserves the complete capture sequences without separating individual actions into short clips.
The recordings retain their continuous capture… See the full description on the dataset page: https://huggingface.co/datasets/Appleartsstudios/Full_Length_Motion_Capture_Dataset.TopoSlots-MotionData
TopoSlots Motion Data
Unified multi-skeleton 3D motion dataset for TopoSlots: topology-agnostic per-slot motion tokenization with foundation alignment.
Paper target: NeurIPS 2026 / ICLR 2027
Last updated: 2026-03-27
Motions: 24,448 across 7 datasets, 79 skeleton types (6 human + 73 animal species)
Dataset Summary
Dataset
Motions
Joints
Skeleton Type
Text Coverage
Text Quality
Renders
humanml3d
14,449
22
Human (SMPL)
100%
High (human multi-caption)
14,449… See the full description on the dataset page: https://huggingface.co/datasets/Tevior/TopoSlots-MotionData.MotionEdit-BenchMotionEdit-Bench is a benchmark dataset for evaluating image editing models on the task of Motion Image Editing, a novel text-based image editing task that aims at modifying actions, poses, and interactions of subjects and objects in images instead of just static features like color.
You can use huggingface datasets to read our dataset:
from datasets import load_dataset
dataset = load_dataset("elaine1wan/MotionEdit-Bench")["train"]
Gen2Humanoid-HY-Motion-1.0w0rldw3aver360-motion
w0rldw3aver360-motion
6 curated 15-second 360° equirectangular video clips sourced from NASA public-domain media — a redistributable, legally-clean dataset for Gaussian splatting, dynamic scene reconstruction, and 360-video model training.
Why this exists
Hugging Face hosts several large "360 video" collections, but none carry a usable license (YouTube rips / gated robotics data). This dataset fills that gap with verified public-domain U.S. Government footage… See the full description on the dataset page: https://huggingface.co/datasets/TheMindExpansionNetwork/w0rldw3aver360-motion.maniskill-pickcube-motionplanning-lerobot-v3
ManiSkill PickCube motion-planning — verified LeRobot v3 reference subset
Need to turn your own raw robotics data into LeRobot v3.0? Start a free conversion →
Community conversion produced by ViaCatalyst BYOD. This repository is not an official upstream release and is not affiliated with the ManiSkill authors.
This is a compact, provenance-complete conversion of the first 10 trajectories in the pinned ManiSkill PickCube-v1 motion-planning HDF5 file. It is a reproducible… See the full description on the dataset page: https://huggingface.co/datasets/ViaCatalyst/maniskill-pickcube-motionplanning-lerobot-v3.paramount_motionmotionatlas-bench
MotionAtlas-Bench v1
MotionAtlas-Bench v1 is a video multiple-choice benchmark for motion and target-entity understanding. This public release contains MCQ records, answer keys, media, and target-object masks needed to reproduce the visual grounding settings.
Resources
Paper: https://arxiv.org/abs/2606.29531
Project page: https://kagura-0001.github.io/projects/MotionAtlas/
Code: https://github.com/Kagura-0001/MotionAtlas
MotionAtlas-Data:… See the full description on the dataset page: https://huggingface.co/datasets/maxLWSv2/motionatlas-bench.text-2-video-human-preferences-motion-v2-large
Human Preferences for AI-Generated Video: Motion Quality v2 (large)
115,732 pairwise human preference labels comparing 4 frontier video generation models on human motion across 3 quality dimensions, collected from real annotators via Datapoint AI.
This is an expanded version of the motion quality dataset with 417 unique prompts (up from 60) and 11 motion categories (up from 6).
Why This Dataset
Video generation models are improving fast, but evaluating human motion… See the full description on the dataset page: https://huggingface.co/datasets/datapointai/text-2-video-human-preferences-motion-v2-large.