Researchers from the University of Maryland and Johns Hopkins University have teamed up to create an efficient object detection model.
The device, dubbed Squeezed Edge YOLO, was designed to run on tiny computing platforms. As the name implies, the model was compressed to a minuscule kilobyte size, significantly increasing energy efficiency compared to traditional YOLO models that have been optimized for edge-based machine learning.
To succeed, the researchers focused on optimizing their model for the GAP8 hardware architecture, which consists of a main microcontroller, an additional eight-core processor, and a number of hardware gas pedals. They worked on the EdgeYOLO model, on reducing the size of the input images.
The new algorithm was tested on a pair of edge computing platforms - an AI-deck with a GAP8 microcontroller and an NVIDIA Jetson Nano with 4GB of RAM. After training the Squeezed Edge YOLO model on more than 8,000 images, the object detection capabilities were evaluated. Compared to EdgeYOLO, the new system was 3.3 times faster while consuming 76% less power. In addition, the Squeezed Edge YOLO is 8 times smaller than EdgeYOLO.
The object detection capabilities of the new model are not significantly different from larger models. This combination of accuracy and efficiency could enable Squeezed Edge YOLO to be used in a wide range of autonomous vehicles in the future.
To succeed, the researchers focused on optimizing their model for the GAP8 hardware architecture, which consists of a main microcontroller, an additional eight-core processor, and a number of hardware gas pedals. They worked on the EdgeYOLO model, on reducing the size of the input images.
The new algorithm was tested on a pair of edge computing platforms - an AI-deck with a GAP8 microcontroller and an NVIDIA Jetson Nano with 4GB of RAM. After training the Squeezed Edge YOLO model on more than 8,000 images, the object detection capabilities were evaluated. Compared to EdgeYOLO, the new system was 3.3 times faster while consuming 76% less power. In addition, the Squeezed Edge YOLO is 8 times smaller than EdgeYOLO.
The object detection capabilities of the new model are not significantly different from larger models. This combination of accuracy and efficiency could enable Squeezed Edge YOLO to be used in a wide range of autonomous vehicles in the future.