AnthroNet: Conditional Generation of Humans via Anthropometrics

Francesco Picetti,Shrinath Deshpande,Jonathan Leban,Soroosh Shahtalebi,Jay Patel,Peifeng Jing,Chunpu Wang,Charles Metze III,Cameron Sun,Cera Laidlaw,James Warren,Kathy Huynh,River Page,Jonathan Hogins,Adam Crespi,Sujoy Ganguly,Salehe Erfanian Ebadi

DOI: https://doi.org/10.48550/arXiv.2309.03812

2023-09-08

Abstract:We present a novel human body model formulated by an extensive set of anthropocentric measurements, which is capable of generating a wide range of human body shapes and poses. The proposed model enables direct modeling of specific human identities through a deep generative architecture, which can produce humans in any arbitrary pose. It is the first of its kind to have been trained end-to-end using only synthetically generated data, which not only provides highly accurate human mesh representations but also allows for precise anthropometry of the body. Moreover, using a highly diverse animation library, we articulated our synthetic humans' body and hands to maximize the diversity of the learnable priors for model training. Our model was trained on a dataset of $100k$ procedurally-generated posed human meshes and their corresponding anthropometric measurements. Our synthetic data generator can be used to generate millions of unique human identities and poses for non-commercial academic research purposes.

Computer Vision and Pattern Recognition,Artificial Intelligence,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop a new human body model capable of generating a wide range of human body shapes and postures. Specifically, the paper introduces a new method named AnthroNet, which conditions the generation of human body models through a large amount of anthropometric measurements. This method can directly model specific human identities and can generate human body models in any posture. Different from previous methods, AnthroNet is the first to achieve end - to - end training using only synthetic data, which not only provides a high - precision human body mesh representation but also allows for accurate anthropometric measurements of the body. In addition, by using a highly diverse animation library, the researchers adjusted the limb and hand movements of the synthetic human body to maximize the diversity of learnable priors in model training.

The main contributions of the paper include:
- Providing a new human body shape model with rich expressions and high resolution, which is conditionally generated based on anthropometric data.
- Proposing two efficient registration pipelines, enabling bidirectional conversion between SMPL and SMPL - X meshes and AnthroNet meshes.
- Releasing a synthetic data generator capable of generating high - resolution meshes and their associated metadata for non - commercial academic research use.

In summary, this research aims to improve the generation ability and diversity of human body models by introducing dense anthropometric data as a condition, while reducing the dependence on expensive 3D scanning devices.