Cameras have become available in many embedded and mobile systems, including vehicles, smartphones, wearable devices, and aerial robots. With the advent
of cameras in these systems comes the possibility of detecting objects in the resulting images using the on-board computers. Energy is limited in mobile systems, however, so for
this possibility to become a viable opportunity, energy usage
must be conservative. The Low-Power Image Recognition
Challenge (LPIRC) is the only competition integrating image
recognition with low power. LPIRC has been held annually
since 2015 as an on-site competition. To encourage innovation, LPIRC has no restriction on hardware or software platforms: the only requirement is that a solution be able to use
HTTP to communicate with the referee system to retrieve
images and report answers. Each team has 10 minutes to recognize the objects in 5,000 (year 2015) or 20,000 (years 2016
and 2017) images. The score is the ratio of recognition accuracy (measured by mean average precision, MAP) and the
total amount of energy consumption (measured by watthour). Team are first trained with images from ImageNet.
They are then given test images during the competition. Each
image may contain one or several objects (such as person,
dog, airplane, and table) that belong to 200 predefined categories. To successfully recognize an object in an image, a
computer program must correctly determine the category
and mark a bounding box around the object. In the first three
years of LPIRC, the champions’ scores improved significantly, to a ratio of 6. 56, as shown in table 1. All teams are from
academe; some teams have sponsorships from industry.
Yung-Hsiang Lu, Alexander C Berg, Yiran Chen
n The Low-Power Image Recognition
Challenge (LPIRC) has been held annually since 2015. This article summarizes the competition advancements
made over the past three years.