Video logo detection by Deep-Transfer Active Learning

Brand logo detection is a special aspect of machine vision. However, Video benchmarks are scarce in the public domain. We exploit power deep convolutional neural network (DCNN) and leverage established datasets related to existing applications develop deep-transfer active-learning (DTAL) algorithm select most valuable samples so that smallest number possible needs be labeled achieve maximum performance improvements for video object model training. By exploiting shared feature space between static through transfer learning based on highly adaptable DCNN features, DTAL implements diversity-based active informative from sequence similar image frames detection. successfully apply new implement live streaming sports videos as well pedestrian face data. show better method than state-of-the-art deep-learning-based techniques. also contribute one largest video-based resources, Sports Match Logo (SMVL) dataset, facilitate general research using transfer-