/half_body_pose_estimation

Primary LanguagePython

Half Body Pose Estimation

Introduction

This is a half body pose estimation project, which can detect four joints(right/left shoulder, neck, head) and three limbs(neck->left shoulder, neck->right shoulder, neck->head).

The algorithm adopted in this project is Realtime Multi-person 2D Pose Estimation using Part Affinity Fields .

Dependencies

pytorch=0.4.1

tensorboadX=1.4

numpy=1.14.3

Files

model: network(CPM and PAF model )

train: train code

inference: inference code

dataset: create input image and labels from AIChanllenge

pose_decode: compute and plot joints and limbs from net output

[1] Realtime Multi-person 2D Pose Estimation using Part Affinity

[2] https://challenger.ai/dataset/keypoint