google/GNM

An open ecosystem of parametric human models and perception stacks, starting with GNM Head.

What it solves

GNM provides a high-fidelity, parametric 3D human model to solve the challenge of accurately representing human geometry and appearance in computer vision and generative AI. It aims to be the most complete 3D parametric human model available, offering a way to disentangle control over identity, expression, and pose.

How it works

The project uses a family of parametric statistical human models. The first release, GNM Head, focuses on the human head and face, providing a statistical model that allows for fine-grained control over identity, expressions, and head pose. It also includes internal anatomy such as eyeballs, teeth, and tongue. The system is designed for flexibility, supporting multiple backends including NumPy, JAX, PyTorch, and TensorFlow.

Who it’s for

Researchers and developers in computer vision, computer graphics, and generative AI who need precise 3D human representations for their applications.

Highlights

  • Multi-framework support: Compatible with NumPy, JAX, PyTorch, and TensorFlow.
  • High-fidelity head model: Provides detailed 3D geometry for the face and face-related internal anatomy (eyes, teeth, tongue).
  • Disentangled control: Separate control over identity, expression, and pose.
  • Permissive license: Released under Apache 2.0 for both commercial and non-commercial use.

Related

  • Project
    facebookresearch/MHR

    A high-fidelity parametric 3D human body model that uses identity, pose, and facial expression parameters to generate realistic, differentiable human meshes.

  • Project
    naver/anny

    A differentiable PyTorch-based human body mesh model that generates parametric 3D human shapes across all ages and body types.

  • Dispatch
    Google DeepMind Visual Alignment Research

    Google DeepMind has developed a three-step alignment method that reorganizes AI visual representations to match human conceptual hierarchies, improving model robustness and generalization.

  • Project
    dluvizon/deephar

    A multi-task deep learning framework for real-time 2D and 3D human pose estimation and action recognition.

  • Project
    deepinsight/insightface

    An open-source 2D and 3D face analysis toolbox providing state-of-the-art algorithms for face detection, recognition, and alignment.