A quick tutorial on multi-pose estimation with OpenCV, Tensorflow and MoveNet lightning.
**Also available on Kaggle
){kind=icon}
I found multipose estimation Notebooks and codes not so explicit or even understandable for pure beginners. Moreover, most of the available tutorials focus on single-pose estimation, with only one instance (human). As a result, the idea of writing my own tutorial naturally came to me. After some research and a bit of styling, code cleaning, presentation...I finally made it public.
- Model type : MoveNet
- Pose estimation method : multipose, bottom-up
- Keypoint count : 17
- Test script
- Bat file for a sample test : requires the model path to be ../Model/TFLite/lite-model_movenet_multipose_lightning_tflite_float16_1.tflite. You can download it directly from TFHub
Args :
Requires Tensorflow Hub and a Kaggle environment. However, feel free to adapt to Notebook to your local setup
- Deep Learning and calculations : Tensorflow 2.x, NumPy, Tensorflow Docs
- Computer graphics/vision : OpenCV
- Display : IPython, Matplotlib
- Image/video writer : image io
- Tensorflow official tutorial