Edge Impulse offers up FOMO algorithm, enabling object detection for microcontrollers
Edge Impulse, a provider of development tools for machine learning (ML) models, has offered up a novel approach that promises to bring object detection capabilities to microcontrollers. The technology could enable a greater range of power-sensitive IoT devices to perform object detection activities such as monitoring of manufacturing processes.
The model, cleverly dubbed “Faster Objects, More Objects” (FOMO), is a new algorithm that enables microcontrollers to perform real-time object detection, tracking and counting in under 200K of RAM. That’s an amount of memory not much larger than the first personal computers used to ship with — back in the 1980’s.
FOMO is also claimed to be 30 times faster than MobileNet SSD, according to Edge Impulse. Researchers said test results showed the algorithm capable of detecting objects in video streams of around 30 fps on the Arduino Nicla Vision (which uses a Cortex-M7 MCU) using 245K RAM.
According to a blog post from researchers, FOMO represents a middle ground between common image processing problems: image classification and object detection.
“Image classification takes an image as an input and outputs what type of object is in the image. This technique works great, even on microcontrollers, as long as we only need to detect a single object in the image,” the authors wrote.
“Object detection takes an image and outputs information about the class and number of objects, position, and size in the image. Since object detection models are making a more complex decision than object classification models they are often larger (in parameters) and require more data to train. This is why we hardly see any of these models running on microcontrollers,” they noted.
The FOMO model represents a simplified type of object detection that is suitable for instances “Where the position of the objects in the image is needed but when a large or complex model cannot be used due to resource constraints on the device,” they wrote.
There are tradeoffs between FOMO and other approaches — the model performs better when the objects are of similar size and not placed to close to each other.
The news comes months after Edge Impulse raised $15 million in a Series A investment round aimed at democratizing machine learning development for edge devices.
For background on edge AI, you can read EdgeIR’s article on what edge AI is and what it is used for here.
Article Topics
AI/ML | algorithm | Edge Impulse | FOMO | manufacturing | object detection | YOLO
Comments