hci-mkim/pose_project

Human Pose Estimation Project

Extracting 2D Joint positions and angles for pose estimation with PyTorch.

Installation

Prerequisites

• Install PyTorch >= 1.2 with torchvision >= 0.4

Option 1: Install using pip

pip install git+https://github.com/prasunroy/rcnnpose-pytorch.git

Option 2: Install from source

git clone https://github.com/hci-mkim/rcnnpose-pytorch.git
cd pose_project
python setup.py install

Running examples

Option 1: only for pose

locate video in folder [1]vid2img

• images will be sliced and saved in folder [2]inputs

cd examples_onlypose
!python vid2img.py

[2]inputs -> [3]output_keypoints

!python custom_image_demo.py

Option 2: background subtraction(yolact++) and then pose

[1]vid2img -> [2]inputs

• locate your video([1]vid2img) and then images will be sliced and saved in folder [2]inputs

cd examples
!python vid2img.py

[2]inputs -> [3]outputs_background_removed

!cd yolact
# run yolact
!python eval.py --trained_model=weights/yolact_resnet50_54_800000.pth --score_threshold=0.15 --top_k=1 --images="../[2]inputs":"../[3]outputs_background_removed"
!cd ..

[3]outputs_background_removed -> [4]outputs_keypoint

!python custom_image_demo.py

Angle.csv results

joint coordinates.csv results