CS5720 - Week 5
Slide 81 of 100

Famous CNN Architectures Timeline

The Evolution of Convolutional Neural Networks

Over the past decades, CNN architectures have evolved dramatically, each breakthrough building upon previous innovations. Let's explore this fascinating journey through the most influential architectures that shaped modern computer vision.
💡 Each architecture addressed specific limitations of its predecessors, pushing the boundaries of what's possible in image recognition and beyond.

CNN Architecture Evolution

1998
LeNet-5
The grandfather of modern CNNs, designed for digit recognition
~60K params 7 layers
2012
AlexNet
Sparked the deep learning revolution with ImageNet victory
60M params 8 layers
2014
VGGNet
Showed that deeper networks with smaller filters work better
138M params 16-19 layers
2014
GoogLeNet/Inception
Introduced inception modules for efficient multi-scale processing
7M params 22 layers
2015
ResNet
Revolutionary skip connections enabled very deep networks
25.6M params 50-152 layers
2017
DenseNet
Connected every layer to every other layer for maximum information flow
8M params 121+ layers
2017
MobileNet
Efficient architecture for mobile and embedded vision applications
4.2M params 28 layers
2019
EfficientNet
Balanced network scaling for optimal accuracy vs efficiency
5.3M-66M params B0-B7 variants
Prepared by Dr. Gorkem Kar