Realtime Multi-Person Pose Estimation

This is a keras version of Realtime Multi-Person Pose Estimation project

Introduction

Code repo for reproducing 2017 CVPR paper using keras.

Requirements

Linux/Ubuntu system if not, try on Google Colab. Works like charm.!

Contents

0. [intro to GoogleColab] 1.Training 2.Testing

Google Colab

• Go to google colab and create new ipynb file.
• !git clone {this repository clone}
• go to dataset folder
• !wget http://images.cocodataset.org/zips/test2017.zip
• !wget http://images.cocodataset.org/zips/val2017.zip
• !wget http://images.cocodataset.org/zips/train2017.zip
• enter these commands one by one. then unzip each of them using !unzip {filename}
• create annotations folder and upload these files from the given link https://drive.google.com/open?id=1Oa4FKj4xwOz_44Psi059Y3-on4Ebz0uv
• DONE! YOUR DATASET IS READY

Training steps

• Go to the "training" folder cd ../../../training.
• Optionally, you can set the number of processes used to generate samples in parallel dataset.py -> find the line df = PrefetchDataZMQ(df, nr_proc=4)
• Run the command in terminal python train_pose.py You can change no.of epochs in train_pose.py file
• This will take some time depending upon your pc/ no.of epochs

Testing steps

• Convert caffe model to keras model or download already converted keras model https://www.dropbox.com/s/llpxd14is7gyj0z/model.h5
• Run the notebook demo.ipynb.
• python demo_image.py --image sample_images/ski.jpg to run the picture demo. Result will be stored in the file result.png. You can use any image file as an input.

Related repository

• CVPR'16, Convolutional Pose Machines.
• CVPR'17, Realtime Multi-Person Pose Estimation.

Thanks to our mentor Dr. Tapas Badal for continuos encouragement.