STAR

Sparse Trained Articulated Human Body Regressor

This project is modified from STAR to provide a handy way of generating body meshes.

Installation

1. Install PyTorch

This package is build on top of pytorch.

See pytorch/get-started for the installation guide.

2. Install STAR

This project is managed via poetry and can be easily installed via pip:

pip install git+https://github.com/digitaldomain-ai-research/star.git

If you are using poetry, run the following cmd instead:

poetry add git+https://github.com/digitaldomain-ai-research/star.git

Quick start

The data generator is modified from STAR. The value of device can be cpu or cuda, but it needs a gpu to do the calculation either way.

from star.star import STAR
import torch
import trimesh

device = "cuda"

star = STAR(gender='female')

poses = torch.zeros((1, 72), device=device)
betas = torch.randn((1, 10), device=device) * 1.25
trans = torch.zeros((1, 3), device=device)

vertices = star(poses, betas, trans).detach().squeeze().cpu()
mesh = trimesh.Trimesh(vertices, star.faces)

mesh.show()