ICME 2016 Image Recognition Grand Challenge

This paper summarizes the MSR Image Recognition Challenge (IRC) running with ICME 2016 Grand Challenges. Since 2013, Microsoft Research has hosted a series of IRCs to motivate the academic and industrial community to solve real-world large-scale image retrieval and recognition problems. This IRC in ICME 2016 continually leveraged the Clickture dataset [1], a large-scale real-world image click data consisting of 40M web images, and a derived subset of 95K dog-related images for the challenge of dog breed recognition. To conduct fair and efficient evaluation, and make the recognition result more reproducible and accessible, the contest runs on an open platform, Prajna Hub, which can help convert a research algorithm into an online service with minimal effort of just a few hours. As part of the ICME 2016 Grand Challenges, more than 30 teams participated this year's MSR IRC and 10 teams successfully finished the task. More details of data, system, metrics, process, and result are described in this paper.