This software implements a human pose regression method based on the Soft-argmax approach, as described in the following paper:

Human Pose Regression by Combining Indirect Part Detection and Contextual Information (link)

The network is implemented using Keras of top of TensorFlow and Python 3.

We provide a code for live demonstration using video frames captured by a webcan. Small changes in the code may be required for hardware compatibility.

The software requires the following packges:

• numpy
• scipy
• keras (2.0 or higher)
• tensorflow (with GPU is better, but is not required)
• pygame (1.9 or higher, only for demonstration)
• matplotlib (only for demonstration)

If any part of this source code or the pre-trained weights are useful for you, please cite the paper:

@article{LUVIZON201915,
title = "Human pose regression by combining indirect part detection and contextual information",
author = "Diogo C. Luvizon and Hedi Tabia and David Picard",
journal = "Computers \& Graphics",
volume = "85",
pages = "15 - 22",
year = "2019",
issn = "0097-8493",
doi = "https://doi.org/10.1016/j.cag.2019.09.002",
}

The source code and the weights are given under the MIT License.