person_to_yolov4

This script uses Tensorflow and SSD Mobilenet V2 (COCO) to recognize people and then update a yolov4 dataset with captions and bboxes that are appropriate to be loaded into DarkMark.

How to use

1. Download the repository

cd ~/koshee
git clone git@github.com:Koshee-AI/person_to_yolov4.git

2. Download the protect-scripts repository (required to link to captions list)

cd ~/koshee
git clone git@github.com:Koshee-AI/protect-scripts.git

3. Install all requirements

cd ~/koshee/person_to_yolov4
./run.sh -h

4. Run person_to_yolov4.py

Get Help

python ~/koshee/person_to_yolov4/person_to_yolov4.py -h

Typical Usage

cd new_directory_with_images_dir
python ~/koshee/person_to_yolov4/person_to_yolov4.py -g 20 "a person moving an item under their shirt"

The above would label images inside of new_directory_with_images_dir/images with a 20% bigger box than default person selection and label that box with the caption a person moving an item under their shirt