metadata
title: MultiView-4D-Human-Motion-Dataset
emoji: 🐨
colorFrom: pink
colorTo: red
sdk: static
pinned: false
license: mit
short_description: Large-scale 1080P multi-view continuous 4D human motion data
Multi-View Continuous 4D Human Motion Dataset
Large-Scale 1080P Multi-View Human Motion Data for AI Research & Commercial Applications
10,000+
Total Hours
2,000+
Real-world Scenes
6,000+
Subject Instances
Dataset Scale
800+Multi-Person Interactions (Hours) |
2,200+Dual-Person Actions (Hours) |
7,000+Single-Person Actions (Hours) |
📂 Dataset Structure
We provide modularly selectable annotation data covering human body parameters ranging from video-level to frame-level.
├── Dataset_Root/
│ ├── Videos/ # raw video data
│ │ ├── Duration/ # duration
│ │ ├── Scene_01/ # a complete acquisition unit
│ │ │ ├── Cam_01.mp4
│ │ │ └── ... (6 views)
│ ├── Content/ # scene content description
│ │ ├── Environment/ # environment
│ │ ├── Scene_Type/ # scene
│ │ ├── People_Count/ # number of people
│ │ ├── Action_Category/ # action
│ ├── VGGT/ # 3D Point Cloud
│ ├── Annotations/ # SMPL Parameters
│ │ ├── Scene_01.json
│ │ │ ├── "frames": [
│ │ │ │ ├──{
│ │ │ │ │ ├── "frame_id": 0,
│ │ │ │ │ ├── "smpl_params": {
│ │ │ │ │ │ ├── "global_orient": [rx, ry, rz],
│ │ │ │ │ │ ├── "body_pose": [[j1_rx, j1_ry, j1_rz],...], # 23 joints in total
│ │ │ │ │ │ ├── "transl": [tx, ty, tz],
│ │ │ │ │ ├── }
│ │ │ │ ├──}
│ │ │ ├──]
💡 Tip： Required fields can be freely chosen based on actual needs: multi-camera raw video data, environment, scene type, number of people, action category, duration, point cloud data (.vgt), frame-level human body parameters (SMPL).
📊 Dataset Overview
- multi-person actions： 800+ 小时
- two-person actions： 2,200+ 小时
- single-person actions： 7,000+ 小时
- number of real-world scenes： 2,000+
- Shooting participant person-times： 6,000+
- total actual shooting duration： 10,000+ 小时
- video quality： Full set of 1080P HD videos, with the longest single video exceeding 60 minutes.
- acquisition method： Multi-view synchronous acquisition, with stable time-series alignment.