One of the things that Sighthound has historically done well is finding faces. So many customer use cases require us to find human faces that we’ve put a lot of research hours into it. And so when medical companies came to us and said they had HIPAA concerns about faces in videos, we turned our attention to blurring the faces to protect privacy.
Our redaction capabilities have now been built into some of the best privacy products in the world, and we hear time and time again that there is nothing better on the market. But with the advent of GDPR, we wanted to make sure that our technology adapts well to the use cases it brings with it, notably anonymizing surveillance video and smart city street scene captures. And so we’re happy to announce that the new version, Redactor 3.0, blows away its predecessor in every way.
The new Redactor adds more classes of automated object detection to reduce manual intervention in video redaction. In addition to ﬁnding faces, Redactor 3.0 detects people, cars, trucks, buses, motorcycles and license plates for automatic tracking and blurring. Accuracy in ﬁnding faces improves by 29%. New computer vision models are dramatically better at ﬁnding small faces, particularly in surveillance videos. Finally, Redactor 3.0 is optimized for 16 core CPU machines, has Intel acceleration, and nVidia GPU support.
In addition to automating more detection classes, Redactor 3.0 improves face detection by incorporating the latest small face model from Sighthound IO computer vision platform. Better at distant faces, and better at proﬁle faces, the result is a more thorough identiﬁcation of all subject faces in the scene, and less time using the manual redaction and tracking tools. See the following images as an example.
Redactor 3.0. Now available for Linux and Windows. The state-of-the-art just got better. If you’d like to hear more, or try Redactor 3.0 for yourself, click here.