Command Palette
Search for a command to run...
Bianco Simone Buzzelli Marco Mazzini Davide Schettini Raimondo
Abstract
In this paper we propose a method for logo recognition using deep learning.Our recognition pipeline is composed of a logo region proposal followed by aConvolutional Neural Network (CNN) specifically trained for logoclassification, even if they are not precisely localized. Experiments arecarried out on the FlickrLogos-32 database, and we evaluate the effect onrecognition performance of synthetic versus real data augmentation, and imagepre-processing. Moreover, we systematically investigate the benefits ofdifferent training choices such as class-balancing, sample-weighting andexplicit modeling the background class (i.e. no-logo regions). Experimentalresults confirm the feasibility of the proposed method, that outperforms themethods in the state of the art.
Benchmarks
| Benchmark | Methodology | Metrics |
|---|---|---|
| image-classification-on-flickrlogos-32 | TC-VII (with outside data) | Accuracy: 96.0 |
| image-classification-on-flickrlogos-32 | TC-VII (without outside data) | Accuracy: 91.7 |
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.