This article is part of our coverage of the latest inAI research.
Mostobject-detection deep learningmodels have memory and computation requirements that are beyond the capacity of small processors.
It’s free, every week, in your inbox.
Object detection models are much more complex than image classification networks and require more memory.
But with the current state-of-the-art models you could only do image classification on microcontrollers.
Image classification is very useful for many applications.
However, much more can be done.
It was a big nuisance that youre limited to these very basic classification tasks.
But their memory requirements are still large.
Even models designed for edge applications are hard to run on small devices.
Moreover, these models are bad at detecting small objects and they need a lot of data.
For example, YOLOv5 recommendsmore than 10,000 training instancesper object class.
The idea behindFOMOis that not all object-detection applications require the high-precision output that state-of-the-art deep learning models provide.
Instead of detecting bounding boxes, FOMO predicts the objects center.
Detecting centroids is much more compute-efficient than bounding box prediction and requires less data.
Single-shot object detectors are composed of a set ofconvolutional layersthat extract features and severalfully-connected layersthat predict the bounding box.
The convolution layers extract visual features in a hierarchical way.
The first layer detects simple things such as lines and edges in different directions.
FOMO removes the fully connected layers and the last few convolution layers.
The output effectively becomes a heatmap for object types.
There are several key benefits to this approach.
First, FOMO is compatible with existing architectures.
Limits of FOMO
The benefits of FOMO do not come without tradeoffs.
It works best when objects are of the same size.
Its like a grid of equally sized squares, each of which detects one object.
you could read the original articlehere.
