Tryolabs demo featured at Khipu 2023 consisting of a squat counter game running on a Raspberry Pi 4 together with a Coral TPU. The initial code was based on a TensorFlow Lite pose estimation example.

Preview

preview.mp4

How to Install

1. Clone the repository

   git clone git@github.com:tryolabs/squat-wars.git
   cd squat-wars

2. Install poetry

   curl -sSL https://install.python-poetry.org | python -
   

3. Select Python 3.9.10 for your environment. If using pyenv, we suggest you do:

   pyenv shell 3.9.10
   pyenv which python | xargs poetry env use

4. Install dependencies and download models

   poetry install
   sh setup.sh

How to Run

The game can be launched with the following command

poetry run python squat_wars/main.py

Additionally, there are a couple of flags that allow the behavior to be customized:

Argument	Description	Required	Default
--model	Path to the model .tflite file	No	squat_wars/models/movenet_thunder_tpu.tflite
--camera	Camera ID. Set to 0 for webcam	No	0

Note A camera needs to exist for the game to work.

It should also support a resolution of 800x448 as it is the current default. If it doesn't, the resolution can be changed here.

Also, only single pose movenet models are supported at the moment. The name of the supported models are:

• movenet_lightning
• movenet_thunder
• movenet_lightning_tpu
• movenet_thunder_tpu

An example with both flags would be the following

poetry run python squat_wars/main.py --camera 0 --model squat_wars/models/movenet_lightning.tflite

License

Copyright © 2022, Tryolabs. Released under the BSD 3-Clause.