mmlab-pose

Introduction

This repository serves as tools for using the ap10k model and others from the ViTPose

Here, all you will need is the weights and configs downloaded from ViTPose

Setup

Make sure to do the following, as the packages are required to run the keypoints script:

cd mmcv
pip install -e .

cd ..
cd ViTPose
pip install -e .

Splitting the model

Certain models are grouped with several models being contained in one .pth file, ('+' in the .pth file) you will need to split the model in its parts. To split the model, you will need to run the following command:

python ViTPose/tools/split_model.py --source /path/to/model.pth

Usage

To get and inference keypoints from your model, simply update the pose_config and pose_model in the keypoints.py file.

python keypoints.py path/to/img

Note: the keypoints file is easily extensible and is meant to be adapted to any use case. This repos goal is to provide an easy set up for the packages needed for using ViTPose pretrained models.