Deep neural networks in computer vision: how they work, where they are used and what problems arise

If you have a general idea of how computer vision works but are hungry for details, then this article is for you.

Object recognition by a neural network on Nvidia systems

— , , . : , .

, , , .

, - , , .

, .

! deep learning’ — . 30 . — Huawei, Airbus, Bosch, Louis Vuitton, Sahara Force India Formula 1 team.

« » .

— , , .

40- , - . . . — . , , « »: , .

90-, , . 2014 , , . .

. . , . , .

, .

— , .

— , , , .

, . , , , , . , .

— , .

An example of object detection in the photo

... and identification — …

. — — , .

( ). , .

— , — , . - 0 255, .

Bitmap pixels and their brightness in single-byte encoding

. , . — , .

Building a color image at the dawn of photography

Constructing a color image from pixels of red, blue and green colors — ,

, , .

. , .

What the basic task of categorizing an image looks like

: (, ) ( ). , , .

Types of classification problems

, . : , . , , . - , .

, . , . , , , - .

Conditional scheme for solving the classification problem

— , . :

An example of the simplest decision tree

. , , , .

, , - .

( , ) — , .

Examples of features that can be used to categorize photos — ,

?

. , , , , , .

— , «» , ( ), . , , , , :

— , ():

, , , :

, . , , , . , «».

? , , . , . , , (, ) .

Solving the classification problem using a neural network

: , . .

2010 , ImageNet, 10 . , : , .

.

Examples of bases of marked-up images

, , .

, . - . , . ( Transfer Function — ), — , — . . .

An example of sequential application of filters to an image

, . . , — . - .

, , , , - . 60 — .

, 2010- 2015 , , , .

2010- 2015 ( — )

3–3,5%, , . 4–5%.

, , . , .

, , , . , . , . , -, . .

, , , CPU .

Nvidia MythBusters

(GPU), . . , , , .

?

, , , .

, , 3D.

.

, , ,

, , , , .

, ,

, , . , . . , , , « ».

, , , , , , .

Google. . , ( ). , ( ). .

. .

— . , , ? , .

,

. 99%.

.

, , . :

: « , »

, , , . , , . , , .

— , , . .

, , , .

. . . , Amazon , .

, . , .

— , .

, . .

.

.

, , : , .

,

.

.

, , . — , ( «» ). , , . .

, , . , ( , ) , .

( ). — , , , , .

, ( — )

, , : , , . , .

, . , , , . , , , . , . .

— , . , , . , . . , .

— . , . . — . , . , , .

3D- , . : 12 .

, , , , . . , (). , , .

, , , .

, , . , . , - , .

, . , , , , , . , . . , .

— —

, . . . , , ( ).

, , , . , , . fMRI- ( , , ). , . , .

, Amazon Go. . , . . .

Amazon Go (: SounderBruce, CC BY-SA 4.0)

— . , , , , .

Recognition of product display by photo

, . .

— . , — .

3D positioning using a neural network to help an autonomous vehicle navigate through space — 3D ,

, , , , , , . , .

Visual navigation using neural networks that allows an autonomous car to accurately determine the distance to surrounding objects — ,

. , , . . , ImageNet 10 .

, , - . .

— . . , .

-, . , . .

, , , , . , .

, .

-, . : . , .

. , (: spectrum.ieee.org)

, , .

«» (: spectrum.ieee.org)

. , .

. . : , . .

For deception, they use a base printed on a 3D printer, which is pasted over with photographs — 3D- ,

, , .

— . ? , . — . , .

All Articles