Magritte is an open source Machine Learning tool designed for object blurring and redaction in video files.
This efficient and low-computational tool can be used to detect objects (like license plates and faces) and blur them automatically, in near-real-time, while preserving user experience. Engineers are finding the tool is useful for many industries who may want increased privacy assurances when dealing with sensitive footage.
What is Magritte?
Magritte is an open-source Machine Learning (ML) tool which automatically blurs objects in photos and videos. It uses low computational resources to detect arbitrary objects, such as license plates, and applies a blur once they appear on screen.
According to the Google cloud team responsible for this project, Magritte is “an open-source version of an internal project, Magritte, … uses Machine Learning (ML) advances to detect objects using low computational resources, and applies a blur to those objects automatically, as soon as they appear on screen.” the team explains, that “the tool can blur arbitrary objects, like license plates, and more.”
Wow! Video redaction using ML to ID with low resources.
Google has launched their internal privacy tool, they’ve used for some time internally, as part of their Protected Computing initiative, which aims to ensure the safety and privacy of data processing.
Sounds awesome and we can’t wait to use this awesome bit of code!
What are the benefits of using Magritte?
Magritte benefits the general population and the privacy of the public. Journalists, surveillance and security, business, and numerous industries can continue to monitor as necessary while protecting the privacy of the innocent bystanders nearby.
Due to the exceptionally high accuracy of its underlying algorithms over a range of data, engineers can be confident that Magritte will protect intended targets safely and securely.
Video Redaction tool Sample Video
Here’s a perfect example of how we’re using this tool to redact faces from video data. You can see how well Magritte does at identifying the individuals and tracking their bodies despite constant motion.
This sample was very simple to process and required very little resources to redact.
Google is developing these privacy-enhancing technologies, and have become a focus of the US Government’s Office of Science and Technology Policy. In 2019, Google released its differential privacy library on GitHub to make it easier for developers to use PETs in their own applications.
The concept was also featured in a contest run by the US and UK governments which aimed at creating solutions that protect personal data while still allowing Artificial Intelligence models to be trained.
Where can I learn more about Magritte?
To learn more about Magritte, visit the official Github page.
We’re working on gathering more information about Magritte, and several BIG engineers are using this in active projects, so we will have more to report as we learn about this technology.