https://github.com/amogh7joshi/pose-estimation

Human pose estimation using a PyTorch implementation of PoseNet.
https://github.com/amogh7joshi/pose-estimation

human-pose-estimation pose-estimation posenet pytorch

Last synced: 8 months ago
JSON representation

Human pose estimation using a PyTorch implementation of PoseNet.

• Host: GitHub
• URL: https://github.com/amogh7joshi/pose-estimation
• Owner: amogh7joshi
• Created: 2020-12-18T23:04:10.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2020-12-31T21:32:12.000Z (almost 5 years ago)
• Last Synced: 2025-01-02T15:49:54.056Z (9 months ago)
• Topics: human-pose-estimation, pose-estimation, posenet, pytorch
• Language: Python
• Homepage:
• Size: 361 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Pose Estimation

![snowboarder](examples/example.jpg)

This repository contains the source code for neural networks used in human body pose

estimation. 

The neural networks are built off of the tensorflow.js [PoseNet](https://github.com/tensorflow/tfjs-models/tree/master/posenet) model, 

and the repository is configured as such. The model is then converted to a format usable by PyTorch, and used as such.

## Structure

The system functions in three stages: 

1. Download the relevant weight files for the tensorflow.js PoseNet model and format them into the `data/weights` directory.

2. Create the PyTorch MobileNet v1 model in the `architecture` directory, then convert and load the PoseNet weight files into the MobileNet v1 model.

3. Decode the outputs of the model using methods in the `decode.py` script.

4. Finally, draw the keypoints and pose skeleton onto the image using the `util/draw_utils.py` script.

## Credits

The model output decoding and keypoint/skeleton drawing was loosely based off of Ross Wightman's similar inplementation

of PoseNet into PyTorch, accessible [here](https://github.com/rwightman/posenet-python).