Advancements in CNN Architectures for Computer Vision: A Comprehensive Review
Advancements in CNN Architectures for Computer Vision: A Comprehensive Review
Anumol C S
TLDR
An in-depth analysis and comparison of recent advancements in CNN architectures for computer vision applications is presented, providing a thorough examination of 21 state-of-the-art CNN architectures, such as LeNet-5, AlexNet, ZFNet, VGGNet, NiN, GoogleNet, and GhostNet.
Abstract
Over time, Convolutional Neural Networks (CNNs) have established themselves as robust and influential tool in a variety of computer vision tasks, including image classification and object detection. The comprehensive review paper presents an in-depth analysis and comparison of recent advancements in CNN architectures for computer vision applications by providing a thorough examination of 21 state-of-the-art CNN architectures, such as LeNet-5, AlexNet, ZFNet, VGGNet, NiN, GoogleNet (Inception v1), ResNet, DCGAN, Inception v2/v3, SqueezeNet, MobileNet v1, Xception, ResNeXt, DenseNet, ShuffleNet v1, SENet, Inception v4, MobileNet v2, ShuffleNet v2, MobileNet v3, and GhostNet. Each architecture is critically evaluated based on its design principles, architectural components, suitability for various computer vision tasks, applications where these architectures have shown exceptional performance and discussions of their advantages and limitations. This review aims to provide guidance to both researchers and practitioners in the selection of the most suitable architecture for addressing specific computer vision challenges. The insights gleaned from this analysis can contribute to the ongoing research and development of CNN architectures, propelling the advancement of computer vision systems and encouraging the creation of systems that are not only more robust but also more efficient.