Posenet demo example

ENV

Configuration:

• python 3.7.3
• tensorflow 2.0
• numpy 1.17

more info look in requirements.txt

RUN

1. cd examples/lite/examples/posenet/python
2. Download posenet from page https://www.tensorflow.org/lite/models/pose_estimation/overview in path model/. Check model path in pose.py: posenet = posenet.Posenet("model/posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.tflite")
3. python pose.py.

WORK RESULTS

After run program print in console message with body part and its point:

Processing file: images/source/00001.jpg
{'nose': (129, 89), 'leftEye': (132, 82), 'rightEye': (130, 82), 'leftEar': (144, 85), 'rightEar': (146, 84), 'leftShoulder': (152, 116), 'rightShoulder': (158, 114), 'leftElbow': (157, 164), 'rightElbow': (157, 156), 'leftWrist': (123, 192), 'rightWrist': (125, 186), 'leftHip': (188, 180), 'rightHip': (196, 175), 'leftKnee': (182, 239), 'rightKnee': (186, 240), 'leftAnkle': (168, 296), 'rightAnkle': (228, 296)}

In path image/output you can look file same it: Source file in path images/source

Troubles

1. Sometimes net change right and left point - see 0020.jpg
2. If object is invisible, net place find it anytime - see 0020.jpg

License

Apache License 2.0

Author

Prostakov Aleksey, email: aleksey.prostakov@gmail.com