Mocap Datasets

Research and open motion-capture datasets in BVH, FBX and SMPL. Most are non-commercial.

• 100STYLE. Ian Mason's Xsens BVH locomotion set covering 100 walking and running styles, about 4M frames at 60 fps.
• 3DPW Free, non-commercial. MPI 3D Poses in the Wild: 60 outdoor video sequences with SMPL poses recovered from IMUs and a handheld camera.
• ACCAD Open Motion Project. Ohio State ACCAD Vicon mocap library of walking, running, martial arts, gestures, and dance, in BVH, C3D, and ASF/AMC.
  See also: Animation & Rigging → Mocap & Animation Libraries
• AIST++ Dance Motion Dataset. Google 3D dance dataset reconstructed from multi-view AIST video, 5.2 hours and 1,408 sequences across 10 genres with music.
• AMASS Free, non-commercial. Max Planck archive unifying 15+ optical mocap datasets into SMPL/SMPL-X body parameters, 40+ hours and 300+ subjects.
• ARCTIC Free, non-commercial. ETH and MPI dataset of two hands manipulating articulated objects, 2.1M frames with MANO hands, SMPL-X bodies, and 9 views.
• BABEL Free, non-commercial. MPI action label layer over AMASS: 43 hours of mocap with frame-level and sequence-level English action labels.
• Bandai Namco Research Motiondataset. Free motion-capture dataset from Bandai Namco R&D for research and personal use.
• BEAT Free, non-commercial. Body-Expression-Audio-Text dataset of 76 hours of speech gesture mocap from 30 speakers, with audio, text, and emotion labels.
• BEAT2 (EMAGE) Free, non-commercial. BEAT2 dataset from the EMAGE project: 60 hours of SMPL-X body, hand, and FLAME face mocap for co-speech gesture.
• CHI3D Free, non-commercial. IMAR close-interaction mocap of pairs hugging, pushing, and fighting, with contact annotations and GHUM/SMPL-X fits.
• CMU Graphics Lab Motion Capture Database. Carnegie Mellon optical mocap archive of 2,600+ clips across 140+ subjects, searchable by subject and category, in ASF/AMC and C3D.
• CMU Panoptic Studio Dataset Free, non-commercial. Carnegie Mellon dome captures from 480 VGA and 31 HD cameras, with 3D body, hand, and face keypoints of social interactions.
• DIP-IMU Free, non-commercial. MPI Deep Inertial Poser dataset of 10 subjects wearing 17 Xsens IMUs, about 90 minutes with SMPL pose ground truth.
• EgoBody Free, non-commercial. ETH egocentric dataset of social interactions captured with HoloLens 2 and Kinects, with SMPL-X fits for both people.
• FaceScape Free, non-commercial. Nanjing University 3D face dataset of 938 subjects with 20 expressions each, topology-unified meshes and 4K textures.
• FineDance Free, non-commercial. Tsinghua dance mocap dataset of 14.6 hours across 22 genres with paired music, SMPL-H body and finger motion.
• Fit3D Free, non-commercial. IMAR fitness mocap dataset of 3M+ frames covering 47 exercises, with marker mocap, GHUM/SMPL-X fits, and multi-view video.
• GENEA Challenge 2023 Dataset Free, non-commercial. GENEA challenge dyadic speech-gesture data based on Talking With Hands, with BVH body motion and audio for both speakers.
• GRAB Free, non-commercial. MPI whole-body grasping mocap of 10 subjects with 51 objects, giving SMPL-X body, MANO hands, and object meshes.
• HDM05 Motion Capture Database Free, non-commercial. Uni Bonn and MPI database of 3+ hours of mocap from 5 actors in 130 action classes, with short labeled clips, in C3D and ASF/AMC.
• Human3.6M Free, non-commercial. IMAR benchmark of 3.6M 3D human poses from 11 actors in 17 scenarios, with synced 4-view video, mocap, and depth scans.
• HumanAct12 Free, non-commercial. Action-to-Motion project dataset of 1,191 Kinect-captured clips in 12 action categories with fine sub-labels, as SMPL poses.
• HumanEva Free, non-commercial. Brown University benchmark syncing multi-camera video with 3D mocap of subjects walking, jogging, and gesturing.
• HumanML3D Free, non-commercial. Text-motion dataset of 14,616 motions from AMASS and HumanAct12 paired with 44,970 English descriptions, 28.6 hours.
• HuMMan Free, non-commercial. SenseTime and NTU multimodal 4D human dataset of 1,000 subjects and 500 actions with color, depth, point clouds, and SMPL fits.
• InterHand2.6M Free, non-commercial. Meta Reality Labs dataset of 2.6M frames of single and interacting hands with 3D joint and MANO annotations from multi-view capture.
• InterHuman (InterGen) Free, non-commercial. Two-person interaction mocap dataset of 7,779 motions and 6.56 hours with text descriptions, from the InterGen project.
• KIT Motion-Language Dataset Free, non-commercial. KIT text-annotated mocap set of 3,911 motions with 6,278 natural-language descriptions, drawn from KIT and CMU databases.
• KIT Whole-Body Human Motion Database Free, non-commercial. Karlsruhe Institute of Technology mocap database of whole-body human and object motion, normalized to the MMM reference model.
• LAFAN1 (Ubisoft La Forge) Free, non-commercial. Ubisoft La Forge BVH mocap set of 77 sequences from 5 subjects, about 4.6 hours of locomotion, fights, dance, and obstacles.
• MANN Dog Motion Capture (AI4Animation) Free, non-commercial. Edinburgh quadruped mocap of a dog walking, trotting, sitting, and jumping, from the Mode-Adaptive Neural Networks paper.
• Motion-X Free, non-commercial. IDEA Research whole-body SMPL-X motion dataset with face and hands, 81.1K clips and 15.6M frames, with text labels.
• Motorica Dance Dataset Free, non-commercial. KTH and Motorica optical mocap of 6 hours of dance in 8 styles, retargeted to one BVH skeleton, with music.
• MPI-INF-3DHP Free, non-commercial. MPI markerless multi-camera 3D pose dataset of 8 actors, 14 views, with green-screen footage for augmentation.
• NUS Motion Capture Database Free, non-commercial. National University of Singapore Vicon mocap clips of locomotion, dance, and martial arts, hosted by SFU, in BVH and C3D.
• Nymeria Free, non-commercial. Meta Reality Labs dataset of 300 hours of daily-life Xsens full-body mocap paired with Project Aria egocentric video and narrations.
• RICH Free, non-commercial. MPI Real scenes, Interaction, Contact and Humans dataset: markerless multi-view capture with SMPL-X bodies, scene scans, and contact labels.
• SFU Motion Capture Database Free, non-commercial. Simon Fraser University mocap clips of dance, martial arts, and locomotion, in BVH, C3D, and ASF/AMC with preview videos.
• SURREAL Free, non-commercial. Inria synthetic dataset of 6M frames rendering SMPL bodies driven by CMU mocap, with depth, normals, and segmentation.
• TED Gesture Dataset Free, non-commercial. ETRI pipeline and prebuilt dataset of 3D upper-body gesture poses extracted from 1,766 TED talk videos with transcripts.
• TotalCapture Free, non-commercial. University of Surrey CVSSP dataset syncing 8-view video, 13 IMUs, and Vicon mocap for 5 subjects performing actions.
• Trinity Speech-Gesture Dataset Free, non-commercial. Trinity College Dublin mocap of spontaneous speech with synced audio and BVH body motion for gesture generation.
• ZEGGS Dataset (Ubisoft ZeroEGGS) Free, non-commercial. Ubisoft La Forge speech gesture set of 67 monologue takes in 19 styles, BVH plus WAV, with FBX characters.