Detectum is a deep learning neural network algorithm that can run on very low power. Xailient created Detectum to fulfill the need for fast deep learning model inference at the sensor Edge.
This work has paid off, as Detectum can now run up to 509x faster compared to other object detectors on the market, all while using less power and memory.
What is Detectum?
At Xailient, we built the world’s smallest and fastest object detector, known as Detectum. By object detection, we mean locating objects of interest by drawing boxes around them in images and video. Detectum fits on extremely tiny devices running on exceptionally low power, particularly battery-operated devices.
Specializing in TinyML computer vision for Edge devices, Xailient uses its patented technology to make embedded-Edge computer vision (CV) accurate, real-time, and cost-effective, solving the most difficult problems in the Enterprise CV lifecycle. This is vital because most CV models that have been optimized to run on tiny Edge devices have lost a lot of accuracy in the optimization process, rendering them useless.
How does Detectum work?
Detectum brings the benefits of real-time automation out of the realm of ideas and into practical application. Detectum offers you the ability to transform your business through:
Data science
Xailient AI is trained to answer your questions. The process starts with training data – examples of what you are looking for. Our Orchestrait CV management platform helps you collect and curate this data.
AI engineering
Automated AI training uses our patented Neural Architecture Search to deliver the optimal AI that meets your performance criteria. No coding or special skills are required.
Deployment
Edge deployment has a lower operating expense and better performance, and Xailient supports the most popular embedded systems. Your trained AI is automatically integrated into an embedded SDK at the click of a button, exposing standard interfaces for easy integration.
You can also download a ready-to-deploy Docker Container with REST APIs. Deploy in the Cloud or on-premise servers, and sleep easy knowing that you have future-proof deployment when your organization migrates to the Edge.
Monitoring and management
Ongoing data collection and over-the-air updates are natively supported, easing the burden of managing your cameras and AI at scale. This is important because data and model-drift are common problems often overlooked in the AI integration pipeline.
Privacy Filter
Our privacy safe data collection technology allows you to continuously collect data while redacting faces. Now you can handle “data drift” and “model drift” and stay in compliance with GDPR and other privacy regulations.
No other company can provide you with custom, embedded AI at the push of a button, and only Xailient has the end-to-end integrated system for managing TinyCV at scale.
We’ve solved the most difficult problems in the Enterprise CV lifecycle, so you can see what matters. Now, you focus on your customers and what makes your business unique.
The challenge of running deep learning under resource constraints
Applications are now demanding high-accuracy, real-time computation and real-time responses on low-power and low-cost devices. A major challenge is to meet this real-time demand by running deep learning computer vision on a computationally limited platform without compromising on accuracy and battery life.
Detectum is the solution
Xailient has proven that Detectum software performs Computer Vision 98.7% more efficiently without losing accuracy.
Detectum object detection, which performs both localization and classification of objects in images and video, has been demonstrated to outperform the industry-leading YOLOv3.
Xailient achieved the same accuracy 76x faster than the cloud baseline and was 8x faster than the edge baseline without the accuracy penalty.
These results are impressive.
However, more recently, Detectum has been proven to run 509x faster than YOLOv5, while maintaining similar accuracy. This result has been verified by the Chamberlain Group in a presentation delivered at the 2021 Embedded Vision Summit.
The development of deep learning CV is advancing at a rapid pace. While it’s getting better in accuracy, industry efforts are increasing its size, thus impacting computational time and cost.
While research is being done to reduce the size of the deep learning models so that they can run on low-power devices, there is a trade-off between speed, accuracy, size, and energy consumption.
Xailient’s Detectum is the answer to this challenge, as it runs 509x faster than YOLOv5 while achieving similar accuracy.
Ultimately, Detectum offers accurate, more efficient AI that runs anywhere, even onboard resource-constrained, tiny Edge devices.
Xailient’s Detectum goes up against 3 cutting-edge face detectors
For this challenge, the top three cutting-edge face detectors (voted and ranked by the developer community) were put up against Xailient’s face detector. The competing face detectors were:
- OpenCV Haar Cascade
- Dlib
- MTCNN
Xailient came out on top, followed by OpenCV Haar Cascade. Dilb came 3rd, and MTCNN came 4th. The results are as follows:
- When comparing Xailient to OpenCV Haar Cascade, we discovered that Xailient ran 16 times faster using only 1/4 of the resources with better qualitative accuracy.
- Comparing Xailient to Dilb, we found that Xailient ran 80 times faster using the same amount of resources with similar qualitative accuracy.
- When Xailient went up against MTCNN, we discovered that Xailient ran 80 times faster using only 1/4 of the resources with similar qualitative accuracy.
Detectum was made for Edge devices
Xailient’s Detectum overcomes the challenges of running computer vision on a computationally limited platform and delivers accurate, high-power, and memory-efficient AI that runs anywhere.
Detectum outperforms YOLOv3 and TinyYOLO. More recently, it has also outperformed YOLOv5 by an impressive margin.
As the face recognition comparison results above demonstrate, Xailient’s Detectum can also accomplish face recognition at record speed, using fewer resources.
Once again, we see that there is no trade-off between speed and accuracy where Detectum is concerned. This means Xailient can easily run its Detectum algorithm on low-powered Edge devices while delivering the most accurate results available.
About the Author
