Pipelines

This repo is created to run and combine different machine learning pipelines with other tasks.

• pipe initializes a model or resource and provides destructor as __del__ to release it.
• If a GPU is detected with CUDA while using YOLOv7, that GPU will be utilized for computations.

setup

• install anaconda

conda activate
conda create -n pipeline python=3.9 -y
conda activate pipeline
pip install -r requirements.txt

Usage

python main.py -p PIPELINE -i INPUT -o OUTPUT [-b BATCHSIZE]

• -p, --pipeline
  • Specifies the pipeline module to use for pose estimation from the pipeline directory.
    • movenet
    • mediapipe
    • face_patch_to_mediapipe
    • yolov7
• -i, --input
  • The path to the input video file.
• -o, --output
  • The path to the output CSV file.
• -b, --batchsize (Optional)
  • Batch size for processing.
  • only applicable to yolov7

models

• Download yolov7 model file to assets/ on GitHub releases if needed~
• Download an untrained auth model to assets/ to test the authentication model

references

• https://github.com/tensorflow/tfjs-models/blob/master/pose-detection/README.md
• https://coral.ai/docs/accelerator/get-started/#requirements