Automatic Logo Detection with Deep Region-based Convolutional Networks

Abstract

Abstract: Deep learning is an important element of data science, which includes statistics and predictive modeling. It is extremely beneficial to data scientists who are tasked with collecting, analyzing and interpreting large amounts of data. It makes this process faster and easier. We present a deep learning system for automatic logo detection in real world images. We base our detector on the popular framework of Faster R-CNNand compare its performance to other models such as Mask R-CNN or RetinaNet. We perform a detailed empirical analysis of various design and architecture choices and show how these can have much higher influence than algorithmic tweaks or popular techniques such as data augmentation. We also provide a systematic detection performance comparison of various models on multiple popular datasets including FlickrLogos-32, TopLogo-10 and recently introduced QMUL-OpenLogo benchmark, which allows for a direct comparison between recently proposed extensions. By careful optimization of the training procedure we were able to achieve significant improvements of the state of the art on all mentioned datasets. We apply our observations to build a detector to detect logos of the Red Bull brand in online media and images. Automatic logo detection and recognition is a challenging computer vision task with many applications. One of the typical purposes of such systems is measurement of brand visibility and prominence in online media or television broadcasts.