autocropper-supervision

This utility uses the YoloV8 model for detecting people, and the YoloV8-face model for detecting faces.

It processes a directory of images, detects the people with faces, auto-crops the images (as a copy) and saves them to a specified directory.

Processing results for 25 images from Unsplash

Dataset

Command

python "main.py" --images_directory "unsplash-images" --results_folder "unsplash-images-results" --crop_ratio "(3,4)" --person_percent_detection_cutoff 0.075 --person_padding_percent 0.02

Pre-Requisites

Git
Python 3.10
Open cmd
Clone the repository
1. git clone https://github.com/djbielejeski/autocropper-supervision
Navigate into the repository
1. cd autocropper-supervision

Activate Environment and Install Dependencies

python -m venv autocropper-supervision-venv
autocropper-supervision-venv\Scripts\activate.bat
pip install -r requirements.txt

Run

python "main.py" --images_directory "C:/Images"

Cleanup

cmd> deactivate

All Arguments

python "main.py" --images_directory "C:\\images" --results_folder "C:\\results" --crop_ratio "(3,4)" --person_percent_detection_cutoff 0.075 --person_padding_percent 0.02

Command Line Parameters

Command	Type	Example	Description
`--images_directory`	string	`"C:\\images"`	The path the images folder to process
`--results_folder`	string	`"C:\\results"`	Optional Defaults to `\results`
`--crop_ratio`	tuple	`(3,4)`	Optional Defaults to `(3,4)`. Available Options: `(1,1)`, `(2,3)`, or `(3,4)`
`--person_percent_detection_cutoff`	float	`0.075`	Optional Defaults to `0.075`. Percentage of the overall image that must contain a person to be detected. Useful for removing background people from the results
`--person_padding_percent`	float	`0.02`	Optional Defaults to `0.02`. Percent of the detected person to add padding for. Example: Detected person is `512x512px`, will add `512px * 0.02 => 10px` of padding.

djbielejeski/autocropper-supervision