Recent Advances in Logo Detection Using Machine Learning Paradigms: Theory and Practice (Intelligent Systems Reference Library, 255)

✍ Scribed by Yen-Wei Chen, Xiang Ruan, Rahul Kumar Jain

• Publisher: Springer
• Year: 2024
• Tongue: English
• Leaves: 128
• Edition: 2024
• Category: Library

✦ Synopsis

This book presents the current trends in deep learning-based object detection framework with a focus on logo detection tasks. It introduces a variety of approaches, including attention mechanisms and domain adaptation for logo detection, and describes recent advancement in object detection frameworks using deep learning. We offer solutions to the major problems such as the lack of training data and the domain-shift issues.

This book provides numerous ways that deep learners can use for logo recognition, including:

• Deep learning-based end-to-end trainable architecture for logo detection
• Weakly supervised logo recognition approach using attention mechanisms
• Anchor-free logo detection framework combining attention mechanisms to precisely locate logos in the real-world images
• Unsupervised logo detection that takes into account domain-shift issues from synthetic to real-world images
• Approach for logo detection modeling domain adaption task in the context of weakly supervised learning to overcome the lack of object-level annotation problem.

The merit of our logo recognition technique is demonstrated using experiments, performance evaluation, and feature distribution analysis utilizing different deep learning frameworks.

The book is directed to professors, researchers, practitioners in the field of engineering, computer science, and related fields as well as anyone interested in using deep learning techniques and applications in logo and various object detection tasks.

✦ Table of Contents

Preface
Contents
About the Authors
1 Deep Convolutional Neural Networks
1.1 Deep Learning Frameworks
1.1.1 Core Component and Key Elements of Deep Learning
1.2 Feature Extraction Networks
1.2.1 VGG: Very Deep Convolutional Networks for Large-Scale Image Recognition
1.2.2 Residual Networks
1.2.3 Deep Layer Aggregation Networks
1.2.4 Hourglass Framework
1.3 Object Detection Frameworks: Detection Head
1.3.1 Detection Head Functionality in Object Detection Frameworks
1.3.2 Anchor Box-Based Detection Frameworks
1.3.3 Anchorfree Detection Frameworks
1.4 Summary
References
2 Introduction to Logo Detection
2.1 Logo Detection and Its Applications
2.2 Logo Detection Challenges
2.3 Related Work in Logo Detection
2.3.1 Deep Learning for Logo Detection
2.4 Proposed Approaches for Logo Detection
2.5 Summary
References
3 Weakly Supervised Logo Detection Approach
3.1 Weakly-Supervised Logo Detection Using Image-Level Annotation
3.2 Attention Mechanisms
3.3 Weakly Supervised Logo Detection with Dual-Attention Dilated Residual Network
3.3.1 Feature Extraction Backbone Network
3.3.2 Spatial Attention Mechanism
3.3.3 Channel Attention Mechanism
3.3.4 Gradient-Based Grad-CAM for Localization of Logos
3.3.5 Implementation of Channel and Spatial Attention
3.4 Experiments and Results
3.4.1 Implementation
3.4.2 Dataset
3.4.3 Evolution Measures
3.4.4 Comparison with Different Attention Modules
3.5 Summary
References
4 Anchorfree Logo Detection Framework
4.1 Dual-Attention LogoNet for Logo Detection
4.1.1 Overview of the Logo Detection Framework
4.1.2 Layer-Aggregated Hourglass Style Feature Extraction Network
4.1.3 Attention Modules
4.1.4 Detection Head
4.1.5 Overall Framework of Dual-Attention LogoNet
4.1.6 Lightweight CNNs Network Architecture for Practical Applications
4.1.7 Experiments and Results
4.1.8 Implementation
4.1.9 Evaluation on FlickrLogos-32 Dataset
4.1.10 Evaluation with Lightweight CNNs Method
4.2 Summary
References
5 Mitigating Domain Shift in Logo Detection: An Adversarial Learning-Based Approach
5.1 Domain Shift Problem
5.2 Domain Adaptation for Computer Vision Tasks
5.3 Related Work: Domain Adaptation
5.4 Adaptation Using Anchorfree Object Detector for Logo Detection
5.5 Evaluation with Adversarial-Based Domain Adaptation Using LogoNet
5.5.1 Experiments and Results
5.6 Summary
References
6 Unsupervised Logo Detection with Adversarial Domain Adaptation from Synthetic to Real Images
6.1 Unsupervised Domain Adaptation: Synthetic to Real Logo Detection
6.2 Synthesize Images to Avoid Manual Annotation Task
6.2.1 Synthetic Logo Images
6.3 Domain Alignment Using Entropy Minimization
6.3.1 Entropy Minimization
6.3.2 Entropy Minimization Maps Using Mid-Level Feature from Synthetic to Real Logo Images
6.4 Experiments and Results
6.4.1 Datasets
6.4.2 Implementation Details
6.5 Summary
6.6 Discussion and Future Recommendations
References