Babysister

• Babysister

  • Introduction
  • Weights conversion
  • Usage
  • Demo
    • Select ROIs
    • Run
  • Documentation
    • Generate for local using
  • TODO
  • Credits
  • Copying

Introduction

Objects detection and online tracking api.

---

Weights conversion

Pre-trained darknet weights file can be downloaded here.
Place weights file under directory
babysister/YOLOv3_TensorFlow/data/darknet_weights/ and then run:

$ python convert_weight.py

Converted TensorFlow checkpoint file will be saved to the same directory.
You can also download the converted TensorFlow checkpoint file via
Google Drive link or Github Release

---

Usage

babysister/runner.py: example usage

---

Demo

Download demo videos from here, place them in demo folder.

Select ROIs

$ python select_rois.py demo/TownCentre_720p.mp4 --is-video 1 --save-to demo/rois.csv

Run

$ python demo.py video demo/TownCentre_720p.mp4 \
    --input-size [640,360] \
    --score-thresh 0.25 \
    --valid-classes ["person"] \
    --rois-file demo/rois.csv \
    --save-to demo/result/frames/ \
    --log-file demo/result/log.csv

---

Documentation

ReadTheDocs

Generate for local using

$ cd docs 
$ ./start.sh
$ ./build.sh 

HTML doc will be generated to: docs/_build/html/

---

TODO

• Add docstring.
• Try out other detectors. E.g: M2Det.
• Replace sort with deep_sort.

---

Credits

Awesome works that made this tool possible.
https://github.com/pjreddie/darknet
https://github.com/wizyoung/YOLOv3_TensorFlow
https://github.com/abewley/sort

---

Copying

All my code is MIT licensed. Libraries follow their respective licenses.

---