Review of Deep Learning Algorithms for Image Classification

Arthur Ouaknine
Jan 16, 2018 · 9 min read
Image for post
Image for post

Why do we need image classification?

The ImageNet challenge

Image for post
Image for post
Image for post
Image for post
Example of images in the 2012 ImageNet dataset. Left: Carbonara. Right: English Foxhound. Source: ImageNet

The advent of deep learning

Image for post
Image for post
LeNet-5 architecture for digit recognition. Source: Y. Lecun et al. (1998)
Image for post
Image for post
AlexNet architecture for training with 2 GPUs. Source: A. Krizhevsky et al. (2012)

Going deeper

Inception modules

Image for post
Image for post
Inception module. Source: C. Szegedy et al. (2014)
Image for post
Image for post
GoogLeNet architecture. Source: C. Szegedy et al. (2014)
Image for post
Image for post
Inception module factorization after a nxn convolution. Source: C. Szegedy et al. (2015)
Image for post
Image for post
Inception module factorization application replacing 5x5 convolution by two 3x3 convolutions. Source: C. Szegedy et al. (2015)

Residual learning

Image for post
Image for post
Residual learning block architecture. Source: K. He et al. (2015)
Image for post
Image for post
ResNet architecture. Source: K. He et al. (2015)

The Inception-ResNet

Image for post
Image for post
Architecture of an Inception-resnet-A module. Source: C. Szegedy et al. (2016)
Image for post
Image for post
Inception-ResNet architecture using customized Inception-ResNet modules. Source: C. Szegedy et al. (2016)

Squeeze and Excitation

Image for post
Image for post
Squeeze-Excitation-ResNet module. Source: J. Hu (2017)

Neural Architecture Search

Image for post
Image for post
Architecture of the best convolutional modules learnt with NAS computing the next hidden state using the past one as input. Left: the Normal Cell is the module creating the feature maps. Right: the Reduction Cell is the module reducing the size of the feature maps by a factor of two (it replaces max-pooling layers). Source: B. Zoph et al. (2017)

Progressive Neural Architecture Search

Image for post
Image for post
Cell structure of the best PNAS model. Source: C. Liu et al.(2017)

Conclusion

Image for post
Image for post
Overview of the top-5 error rates on the 2012, 2014, 2015 and 2017 ImageNet challenges.

Zyl Story

📲 An app to bring joy everyday to your friends and family.

Thanks to Camille Guilleminot

Arthur Ouaknine

Written by

PhD Student at Telecom ParisTech and valeo.ai — Deep Learning enthusiast

Zyl Story

Zyl Story

📲 An app to bring joy everyday to your friends and family. 📸 Let’s use all our forgotten photos and videos as daily icebreakers 🍋 Time to #ZestYourLife

Arthur Ouaknine

Written by

PhD Student at Telecom ParisTech and valeo.ai — Deep Learning enthusiast

Zyl Story

Zyl Story

📲 An app to bring joy everyday to your friends and family. 📸 Let’s use all our forgotten photos and videos as daily icebreakers 🍋 Time to #ZestYourLife

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store