Skip to content

Latest commit

 

History

History
76 lines (44 loc) · 1.64 KB

README.md

File metadata and controls

76 lines (44 loc) · 1.64 KB
Raw

MediaPipe human_poseEstimation in PyTorch

#搬砖不易，记得star oo~

Teaser

For work reasons, open source can only go so far

based on https://github.com/zmurez/MediaPipePyTorch

Teaser

feature

  1. Friendly Torch Implementation ,using the GPU the GPU
1.1 Face detector (BlazeFace)

1.2 Face landmarks

1.3 Palm detector

1.4 Hand landmarks

1.5 Iris detector
  1. Add Iris recognition function
  2. Part of the interface for each body gesture is more flexible
  3. 3d Avatar is given with reference to U3d

driver model

install

pip -r install requirements

if your need 3D demo，provide a Unity DEMO

(reference this repo:https://github.com/mmmmmm44/VTuber-Python-Unity)

Run

  1. read args in demo.py ,such as：
python demo.py --source sample\tesla.mp4 --detect_face  --detect_iris  --save_file
  1. First,your need a 3D unity model and open project setting
python demo.py --source sample\tesla.mp4 --detect_face   --connect --debug  --port 5066

Train

Recently will first, collate blazeface magic modified SSD training code,waiting~

TODO

  • add BlazeFace train function
  • imporve pose detect function
  • Add conversion and verification scripts
  • Add multiprocess demo
  • verification performance
  • hand and body 3D model demo

reference

1. https://github.com/zmurez/MediaPipePyTorch
2. https://github.com/hollance/BlazeFace-PyTorch
3. https://github.com/mmmmmm44/VTuber-Python-Unity