Scalable Deep Learning Algorithm for Brand Logo Detection in Images

Brand logo recognition is the task of identifying regions with logo and classifying them. Ithas many applications in different domains such as brand protection and market discovery. Sincethe number of brands is not fixed and changes over time, we propose two-steps algorithm. First,we trained a universal logo detector to find areas where logo lookalike objects are located. Nextto classify found areas we used prototypical neural network as a few-shot recogniser. Proposedend-to-end solution is scalable in terms of number of brands and lightweight. We have testedour proof-of-concept solution on our manually labelled images and FlickrLogos-32 dataset.The aim of this work is to propose and implement a scalable deep learning algorithm forbrand detection in images.To achieve this aim, the following tasks are set:1. Analyze the existing solutions for the detection of the brand’s logo in images.2. Analyze the existing labelled data with the logos of the brands in images.3. Propose a scalable deep learning algorithm to detect brand logos in images.4. Select data for train, validation and test sets.5. Receive and analyze the results of the algorithm evaluation.The object of the research is the problem of detecting brand logos on images. The subject ofthe study is scalable deep learning algorithms for solving the problem of detecting brand logoson images.This research was submitted to ACM Multimedia 2019 international conference. And it ispartially based on the work supported by Samsung Research, Samsung Electronics.It contains 25 pages, 6 chapters, 9 figures, 12 tables and 41 references. Keywords:Logo Detection; Universal Detector; Transfer Learning; Few-shot Learning; Pro-totypical Neural Networks; Deep Learning