Convolutional neural networks (CNNs) are a type of deep neural network used for analyzing visual imagery. CNNs have proven effective for tasks like image recognition and classification. A basic CNN architecture includes convolutional layers that extract features, pooling layers that reduce dimensionality, fully connected layers for classification, and a final output layer. CNNs were inspired by research on the visual cortex in the 1950s and have achieved human-level accuracy on large image datasets like ImageNet thanks to developments like AlexNet in 2012, GoogleNet in 2014, and ResNet in 2015. CNNs are now commonly used for computer vision tasks and applications like style transfer.

1. CNN
Convolutional Neural Networks -an introduction

2. Dense NN vs CNN

3. About
Convolutional Neural Networks (ConvNets or CNNs) are a category of deep, feed-forward artificial
neural networks that have proven very effective in areas such as image recognition and classification.
- Most successful with vision
- Could be used for audio and video signals

4. Basic architecture

5. Elements
- Convolutional layers
- Nonlinearity
- Pooling
- Fully connected layers
- Final layer
- Normalization
- Dropout

6. Why do we need CNN

7. Inspiration
David Hubel and Torsten Wiesel experiments in the early 1950s

8. Line detection
Edge detection
Sobel edge
Laplacian
Laplacian of Gaussian

9. Example

10. Convolution

11. Nonlinearity

12. Pooling
- Max Pooling
- Average Pooling

13. LeNet 1998 LeCun

14. ImageNet Large Scale Visual Recognition Challenge
The ImageNet project is a large visual database
designed for use in visual object recognition
software research. Over 14 million URLs of
images have been hand-annotated by ImageNet
to indicate what objects are pictured; in at least
one million of the images, bounding boxes are
also provided. ImageNet contains over 20
thousand ambiguous categories; a typical
category, such as "balloon" or "strawberry",
contains several hundred images

15. AlexNet 2012 Alex Krizhevsky
top-5 error to 15.3%. The
second place top-5 error rate,
which was not a CNN
variation, was around 26.2%.

16. GoogleNet/Inception Szegedy 2014
The winner of the ILSVRC 2014 competition was
GoogleNet(Inception) from Google. It achieved a top-5 error rate
of 6.67%! This was very close to human level performance
which the organisers of the challenge were now forced to
evaluate. As it turns out, this was actually rather hard to do and
required some human training in order to beat GoogLeNets
accuracy. After a few days of training, the human expert (Andrej
Karpathy) was able to achieve a top-5 error rate of 5.1%.

17. VGG Simonyan and Zisserman 2014
The runner-up at the ILSVRC 2014

18. ResNet He at al 2015

19. DenseNet Huang 2016

20. Comparison

21. GAN

22. Usage

23. Transfer Learning

24. Style transfer

25. https://deepart.io

26. END