Open Source RoboticsRobots, software & learning

GVHMR

Software

Recover human motion in world coordinates from video, with CUDA inference and noncommercial research source terms. [1]

Installation & first result →Official websiteGetting started

For
Human-motion researchers preparing motion estimates for analysis or downstream retargeting [1]
Source rights
Some licenses are unresolved. Includes noncommercial terms. Review licenses
Unknowns
operating system not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware
License
Custom noncommercial research license (noncommercial) · license not fully assessed for model weights, datasets · Details by part

Requirements · Research details and dates

On this page

Overview

GVHMR recovers human motion in world coordinates from video. It supplies inference, training and evaluation code for motion research; a separate downstream step is needed to map those estimates to a robot. Its source license allows educational, research and non-profit use only, so this is a source-available research profile with explicit noncommercial terms. [1] [7]

Is this for you?

Known limitations
  • Source terms restrict use to educational, research and non-profit purposes.
  • The -s option assumes a static camera; default moving-camera estimation uses SimpleVO, not DPVO.
  • External assets retain separate terms.

Editorial assessment from cited documentation; no runtime testing.

[4] [7] [2] · Editorial assessment

What you need

Dependencies
  • Python 3.10
  • PyTorch 2.3.0 with CUDA 12.1
  • Linux x86_64 PyTorch3D wheel in requirements
  • Separately obtained SMPL/SMPL-X and pretrained checkpoints

[2] [3]

Compute for the workload
  • CUDA is used by the demo; minimum inference VRAM is not established in this review.

Editorial assessment from cited documentation; no runtime testing.

[4] · Editorial assessment

Installation & first result

Open the upstream getting-started guide [2]

  • Prepare the Python 3.10 environment and install the pinned source and requirements.
  • Place licensed body models and pretrained assets in the documented checkpoint folders.
  • Run the bundled static-camera tennis demo shown in the overview.

Editorial assessment from cited documentation; no runtime testing.

[1] [2] · Editorial assessment

Installation

Install requirements.txt and the pinned checkout with pip install -e . in the documented Python 3.10 environment; acquire body models and checkpoints separately.

[2]

Hardware, models & data

Inputs and outputs

Video produces saved human-motion estimates and camera/global visualization videos.

[4] [5]

Project notes

Prepare the assets before running the demo

Check out commit ee960bb6e2ea2d381aa97f08e9b71ef320b624b1, a snapshot with no GitHub release listed. The installation guide creates a Python 3.10 environment, then runs pip install -r requirements.txt and pip install -e .. Requirements pin PyTorch 2.3.0 with CUDA 12.1 and a Python 3.10/Linux x86_64 PyTorch3D wheel. The supplied requirements are therefore not a portable CPU or macOS installation recipe. [2] [3] [6]

Obtain SMPL and SMPL-X through their own registration/download routes, and place them under the documented body-model directories. Obtain the separately linked GVHMR and preprocessing checkpoints too; the demo configuration expects inputs/checkpoints/gvhmr/gvhmr_siga24_release.ckpt. The repository license does not settle those assets' permissions. [2] [5]

The bundled static-camera example is python tools/demo/demo.py --video=docs/example_video/tennis.mp4 -s. Use -s only when the camera is static: it skips camera-motion estimation. Although an older README sentence says moving cameras use DPVO, the reviewed implementation and configuration default to SimpleVO. DPVO is an optional --use_dpvo path with additional installation steps. [1] [2] [4] [5]

The default output directory is outputs/demo/<video_name>, with hmr4d_results.pt and camera/global rendering videos. The script uses CUDA; no minimum inference VRAM or motion-accuracy result has been independently established here. [4] [5]

Retraining has a separate data boundary

Upstream provides links to preprocessed training and evaluation data but says original datasets must be acquired from their owners. Its instructions retain the original dataset terms and limit the preprocessed data to research use. The code license also requires distributed modifications to remain open source and noncommercial; commercial use requires contacting the rights holder. No external asset payloads, installation, training or robot tests were performed for this review. [2] [7]

What we checked

Source researched only; no installation, training, benchmark or physical hardware test performed. No affiliate links or reused media.

Not yet established

  • Requirements: operating system, host computer
  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Sep 15, 2026 [6]
Latest release
Not established
Profile updated
Oct 17, 2026
Sources last checked
Oct 17, 2026
Next review
Nov 17, 2026
Reviewed revision
commit snapshot; no GitHub releases listed / ee960bb6e2ea2d381aa97f08e9b71ef320b624b1 · software repository

Source files and licenses

Research-use source with explicit noncommercial scope; model/data rights separately unresolved.

LayerAvailability / license
softwareavailable · Custom noncommercial research license
Repository materials only; external components retain their own terms. [7]
model weightsavailable · License not fully assessed
Download routes documented only; payloads not acquired. Separate body-model/checkpoint/dataset terms unresolved; data instructions specify research use. [2]
datasetsavailable · License not fully assessed
Download routes documented only; payloads not acquired. Separate body-model/checkpoint/dataset terms unresolved; data instructions specify research use. [2]
documentationavailable · Custom noncommercial research license
Repository materials only; external components retain their own terms. [7]

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.7.0-draft.5

Change history