PixelateMe 1.0.0

A Python CLI-Tool and package to pixelate or blur faces in images and videos.

Welcome to PixelateMe 👋

Your Python package for anonymising faces in images and videos

This CLI tool lets you pixelate, blur or remove faces from videos and images. GPU acceleration supported

Original image from unsplash.com by Susan G. Komen 3-Day

Original video from pexels.com by fauxels

⚡️ Quickstart

📥 Install

pip install pixelateme

With GPU support (additionally installs onnxruntime-gpu:

pip install pixelateme[gpu]

▶️ Run

After installation, pip registers a shortcut binary which can be called like this:

pixelateme --mode blur FOLDER_OR_FILES

This will create a new pixelated folder to hold all the pixelated files.

To launch the gui directly open the pixelategui executable from the packages folder of your python installation or:

pixelatemegui

📝 Custom code

To use this package in your own code, you can import the main module. This module exports a run method that accepts all CLI arguments as parameters.

from pixelateme.main import run

run(path=["FILE_PATHS"])

🎯 Features

• Different anonymisation modes: pixelate, blur and color
• GPU acceleration
• Preview of currently processed files
• Face Recognition to blur only certain faces or all faces except certain ones
• ONNXRT and OpenCV run-time backend
• Process multiple files in parallel
• GUI

💻 CLI Arguments

• --suffix: Filename suffix of processed files. Default:
• --output (-o): Output directory for processed files. Default: ./pixelated
• --mode (-m): Mode of anonymization. Default: pixelate
• --threshold (-t): Threshold for detected faces (higher means more confidence). Default: 0.5
• --backend: Desired backend (e.g. opencv or onnxrt). Auto prefers onnxrt and falls back to opencv. Default: auto
• --only-blur-this-faces: Folder containing images of faces (one face per image) that should be considered for anonymisation. All other faces will not be anonymised. Default: None
• --blur-except-this-faces: Folder containing images of faces (one face per image), which should be ignored for anonymization. Default: None
• --ellipse: Uses ellipses as form for anonymization. Default is rectangle
• --blur-strength: Defines how "blurry" a face will be. Only working with --mode "blur". Default: 3
• --pixelate-size: Amount of pixelation effect. The higher the value, the harder it is to recognise the face. Default: 3
• --deepface-similarity: The maximum similarity between two faces. A higher value means that more faces are considered equal. Only working in combination with --blur-except-this-faces or --only-blur-this-faces. Default: 0.4
• --preview: Enable preview of the currently processed image. No preview is default
• --face-recognition-size: Image size to use for face detection. Format: WxH (e.g. 720x480). Default: None
• --maximum-face-recognition-size: Maximum number of pixels of the longest side for face detection. Images larger than this will be downscaled for face recognition. This doesn't affect output resolution. Default: 640
• --multiprocessing: Enable multiprocessing to process files in videos. Useful when anonymizing multiple large videos. Disabled by default
• --parallel-processes: Number of parallel processes. Only works with --multiprocessing enabled. Default: 4

🔧 Settings Overview

Mode		Strength 3 (default)	Strength 6
Pixelate
Blur
Color

👏 Acknowledgements

• deface: Deface was one of my inspirations for this package and the implementation of CenterFace.
• CenterFace: Used for face detection
• DeepFace: Used for face recognition
• ONNX: Face detection backend
• OpenCV: Face detection backend, used for I/O and face anonymisation (blur, pixelation and color)

👥 Author

👤 Marius Butz

• Website: http://marius-butz.de

⭐️ Show your support

• Give a ⭐️ star if this project helped you!
• Create a 🍴 fork and contribute by fixing bugs or adding features