Markerless multi-person motion capture that recovers SMPL-X body, hands and shape from multi-view video.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
MAMMA (Markerless Accurate Multi-person Motion Acquisition) is a research pipeline from the Max Planck Institute for Intelligent Systems (tue.mpg.de domain). It recovers SMPL-X parameters from multi-view RGB video, focusing on two-person interactions, and also handles single-person capture.
The pipeline uses MammaNet, a transformer-based dense landmark estimator (ViT-Base backbone plus a CNN for masks) that predicts 512 2D surface landmarks with uncertainty, visibility and contact probabilities (person-person and floor). Landmarks are matched across views using symmetric epipolar distance and Hungarian assignment, with SAM2 used to track people. SMPL-X is then fitted in several stages, including contact optimization that limits interpenetration.
MammaNet is trained on MammaSyn, a synthetic dataset of 2.5M crops extending BEDLAM, plus BEDLAM data. The authors report lower errors than SMPLify, LookMa and CameraHMR on benchmarks including RICH, Harmony4D, CHI3D and MOYO, and introduce MammaEval-Singles and MammaEval-Dance captured with Vicon. A video creator reports that code has been released, but no repository link or license was in the provided material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Markerless multi-person motion capture from multi-view video, recovering pose, shape, and hands of each person even in close contact. Trained on the Mamma-Syn synthetic dataset of 2.5M crops; code released.
Closest neighbours by meaning, not just by category.
Model
Unified motion-language model for text-to-motion, motion captioning, completion and editing, from HKUST and Tencent Games.
Model
Text-to-human-motion flow-matching model that generates full-resolution motion directly, with training-free joint constraints.
Research
Diffusion framework that learns text-to-3D human motion generation from 2D poses extracted from monocular videos.
Model
Monocular video motion capture that combines flow matching with physics-based RL for plausible 3D human motion.
Model · fal
LoRA adapter from fal for MiniMax H3 that targets realistic people, faces and natural skin in text/image/reference-to-video.
Research
Research method that builds controllable, articulated 3D object models from keypoints, point clouds or depth video.