Revolutionizing Efficiency: Gan-augmented Training for Lightweight Image Recognition

Author:

Korla Jayanthi,Yenni Priyanka,Sunkara Venkata Sai Prasad

Abstract

Abstract

In the world of computer science, image recognition is a vital area that uses complex algorithms to make machines able to understand and classify visual data. The technology has applications in many fields including artificial intelligence, computer vision and others as it offers solutions for automated systems, security and object detection. This paper uses LaPlace GAN for image clarity enhancement and SSD (Single Shot MultiBox Detector) for effective object detection using one shot learning. By employing the strengths of LaPlace GAN, the model concentrates on improvement of intricate details and minimization of noise that leads to significant increase in overall clarity of an image. In addition, integration of SSD allows real time object detection hence the system can identify multiple objects simultaneously within a single step. Through training images from both PASCAL VOC2007 and PASCAL VOC2012 datasets, this research shows improved model performance. For training, proposed framework combines union of PASCAL VOC2007 train and VOC2012 trainval images; validation set is composed of PASCAL VOC2007val images while testing will be done with Pascal’s 2007 test images. The assessment gauge contains the Average Precision (AP) and mean Average Precision (mAP) calculated for all 20 classes that have been annotated in line with the PASCAL VOC object detection evaluation protocol.

Publisher

Springer Science and Business Media LLC

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3