Explosion of Data

Sensors enable the explosion

Visual Data is hard to grasp the contents

Help to search the content of data needs visual technology

Problems facing today: massive amount of data and the challenges of the dark matter

To know the problems help you go on

Neuroscience
神經科學

Cognitive sciences
認知科學

optics
光學

Image processing , Speech, NLP, 

Big Bang of Evolution:543million years, B.C. :  

the beginning of visual processing:simple structure of the world

oriented edges

experiments:  awake but anaesthetized cats

little needle electrode to push electrons through to the skull

primary visual cortex: do a log of visual processing

early: tons and tons of new orleans 

1st stage: back of the brain, the furthest of the eyes, not ear the eyes

the edges define the shape:

Birthday of CV: 1966, MIT Standford, AI lab, 

the beginning of deep learning: David Marr, 1970s Stages of Visual Representation

Goal is to reconstruct 3D model: so we can recognize objects

the first wave of visual recognition algorithms went after the 3D model:

the world is composed of simple shapes like blocks

David Lowe, 1987

Normalized Cut (Shi & Malik, 1997)

Face Detection, Viola & Jones, 2001

the first successful high-level visual recognition algorithms being used by consumer product

the first digital camera that has a face detector Fujifilm 2006

deep learning algorithms try to learn simple features 

focus on features: “SIFT” & Object Recognition, David Lowe, 1999 since hard to describe the whole thing

ML tools like SVM to recognize scene: Spatial Pyramid Matching, Lazebnik, Schmid & Ponce, 2006

Deformable Part Model:Felzenswalb, McAllester, Ramanan, 2009

PASCAL Visual Object Challenge (20 object categories), [Everingham et al. 2006-2012]

www.image-net.org 22K categories and 14M images, 

Deng, Dong, Socher, Li, Li, & Fei-Fei, 2009

The Image Classification Challenge: 1,000 object classes 1,431,167 images

the beginning of deep learning evolution

cool problems:

labeling of the entire scene with perceptual grouping

combining recognition with 3D

CS231n focuses on one of the most important problems of visual recognition – image classification

There is a number of visual recognition problems that are related to image classification, such as object detection, image captioning

Convolutional Neural Network (CNN) has become an important tool for object recognition

Convolutional Neural Network (CNN) is not invented overnight

Pre-requisite
• Proficiency in Python, some high-level familiarity with C/C++
– All class assignments will be in Python (and use numpy), but some of the deep learning libraries we may look at later in the class are written in C++.
– A Python tutorial available on course website
• CollegeCalculus,LinearAlgebra
• Equivalent knowledge of CS229 (Machine Learning)
– We will be formulating cost functions, taking derivatives and performing optimization with gradient descent.