Putting highly advanced Deep Learning algorithms, especially for computer vision applications like Autonomous Vehicles and IoT on edge devices requires special capabilities. In order to operate in a regime of high frame rate, low power consumption and low working memory while keeping the model’s accuracy high is something we have been working on for over the last 3 years. Our mission at Brodmann17 is to open up the world to the benefits of deep learning vision. Our patented, lightweight and robust vision technology is perfect for all edge devices and I am confident by providing the ARM developer community open access to our face detector code will result with new innovation that will clearly show the rest of the world the benefits of our next gen deep learning vision.
Today we are sharing the first model we have created, the first version of our face detection algorithm. It was crafted using exciting new proprietary Neural Networks design patterns, such that the provided model is both highly efficient and accurate. We find it to be a superior alternative to other open source face detectors out there in terms of speed/accuracy, especially if your target application is an edge device.
The model is intended to run on CPU processors (ARM & x86). Moreover, as part of our efforts we have developed our in-house inference engine, such that you can run our library pretty much out of the box without additional installations!
Please refer to our github repository for a detailed setup instructions. Once you’re done you will be able to run our library using C++ or Python on ARMv8-A (aarch64) and x86(64).
A speed benchmark on ARM Cortex A-72 is provided below:
Below you’ll find 2 code snippets (Python & c++) that will help you get started. The following is covered:
- Read an image
- Process the image to detect faces
- Display the results (face bounding boxes)
Getting Started with Python
Getting Started with C++
Please refer to our github repository to obtain a copy of our library and for additional information.
Stay tuned for additional releases of models and code.