Human Shape and Pose Estimation

The HMR (End-to-end recovery of human shape and pose, CVPR 2018 by Angjoo Kanazawa, Michael J. Black, David W. Jacobs, Jitendra Malik) runs with python 2.7 on the linux system.

This version of HMR runs with python 3.6 on Windows.

Requirements

• Anaconda env. of Python 3.6
• TensorFlow tested on version 1.12

Installation

Setup conda environment

1. Make an environment with python 3.6 in Anaconda.
2. Install numpy, scikit-image, pyopengl, py-opencv, bottleneck, tensorflow using Anaconda Packaging.

Install opendr

1. Download opendr from https://github.com/polmorenoc/opendr

In a terminal of the conda environment, go to the opendr folder. It has both opendr and chumpy folders.

2. Correct opendr/renderer.py file

Uncomment line 484 to obtain normals in tn() method.

3. Install both on the conda environment by moving into opendr then,

  pip install .

This procedure installs chumpy (0.76) from PyPI, which does not work with opendr. So,

  pip uninstall chumpy

Move into chumpy then,

  pip install .

This install chumpy (0.66) that works with opendr.

4. ipdb & glfw should also be installed, but they are not in Anaconda Packaging.

  pip install ipdb
  pip install glfw

Opendr can be checked as

  python demo_fit_cube.py

It will pop some windows including cubes. If an error occurs in scipy, a simple modification is required. You search for scipy in site-packages of your conda environment. Edit sputils.py at line 281 in scipy/sparse

new_shape = tuple(operator.index(arg) for arg in args)

changed to

new_shape = tuple(operator.index(int(arg)) for arg in args)

Demo

Finally, move to the folder of this package.

1. Download the pre-trained models

wget https://people.eecs.berkeley.edu/~kanazawa/cachedir/hmr/models.tar.gz && tar -xf models.tar.gz

2. Run the demo

python -m demo --img_path data/coco1.png