Yung-Hsiang Lu - Purdue Univ., West Lafayette, IN Alexander Berg - Univ. of North Carolina, Chapel Hill, NC David Z. Pan - Univ. of Texas at Austin, TX Daniel Hammerstrom - Defense Advanced Research Projects Agency, Arlington, VA David Atienza Alonso - Swiss Federal Institute of Technology, Lausanne, Switzerland Massimo Alioto - National Univ. of Singapore, Singapore
The second Low-Power Image Recognition Challenge (LPIRC) will be held at the 2016 Design Automation Conference in Austin, Texas. Cameras have become available in many embedded and mobile systems, including vehicles, smartphones, wearable devices, and aerial robots. It is desirable to have the capability of detecting and identifying objects in images acquired from such devices. Since energy is limited in mobile systems, efficiency and energy conservation are primordial.
The winners must demonstrate high accuracy in image recognition using the least amount of energy. Each image contains one or several objects. The objects are chosen from different categories. A contestant must identify the objects by their categories as well as marking the objects by bounding boxes. The energy is measured by a high-speed power meter. This is an onsite competition and participants must bring their own systems.
LPIRC 2016 has three tracks. The first two tracks (1) offloading and (2) non-offloading are the same as in LPIRC 2015. The test images are sent through a network. In the third, new track, the test images are shown on a computer screen and the contestants must use cameras to obtain the input images.
The first LPIRC was held at the 2015 Design Automation Conference on June 7, 2015 in San Francisco, California. Thirty-four people registered as ten teams from four different countries. In total, twenty solutions were presented.
A special session at the International Conference on Computer-Aided Design 2015 was devoted to the introduction and description of the first LPIRC. The top two winners of the first LPIRC presented their solutions.