Introduction to DL

The fields of Deep Learning and computer vision exists for quite some times.

• 1963: Larry Roberts creates the block world
• 1966: MIT, a guy proposes to solve vision during a summer school... LOL
• 1970: David Marr writes a book all about vision and how humans detect edges to combine them into objects
• 1979: The generalized cylinder model comes to computer vision
• 1999: SIFT features to recognize objects
• 2001: Face detector make its way in real time computer vision
• 2009: Histogram of oriented gradients
• 2005: HOG get to be known
• 2006: Alexnet beats everyone else in Imagenet !
• Since then: Deep learning is the king in computer vision

Image What ?? ImageNet :)

AlexNet's results in the classification task

Detection in imageNet

Yolo's results in the detection task

Is it just about Classifiying images ?

Not at all, yet, it's easier with images...

 Audio Generation

• Audio1
• Audio2

Image generation from picture

https://turbo.deepart.io/

Image Generation from random noise

Image generation from sketching

https://www.autodraw.com/

Amazing! When are we doing this ??

When you have a good computer

Compute power - GPU

You need a lot of compute power to reach these results...

What about this class then ?

We are going to classify this :

What we'll do in this class :

• Implement a logistic regression
• Implement mutually exclusive classification (multiclass regression)
• Understand a Convolutional Neural Network
• Use Tensorflow to solve an Image classification problem (Cats and Dogs)
• Use Tensorflow to solve an audio classification problem (Cats and Dogs)
• Project: Solve the ESC-50 dataset.

Lets Learn about Neural Networks...

Schema of a neuron

Output of a neuron

Dendrites and Axions

How it really looks like

(Saturated Reconstruction of a Volume of Neocortex)