Evaluation on Huawei Accurate and Fast Mobile Video Annotation Challenge

Abstract

Massive user generated content (UGC) videos are produced each day on the Internet. These videos have become a very important integrant in existing social networking services (SNS). However, unlike professional films, the content of UGC videos is usually unstructured and lacks contextual annotation for management. The motivation behind Huawei Accurate and Fast Mobile Video Annotation Challenge (MoVAC) is to evaluate different algorithms on the generation of local annotation on UGC videos under the same protocol, and to compare them not only in accuracy but also in efficiency. More than 15 teams from different countries have enrolled in this competition, and in the final round 17 submissions with valid result from 6 teams were received. The results show that recent popular deep convolutional neural networks (CNN) could be a potentially good solution to this task.