Deep Learning Data
so
how do we make computers see like us one way is to use deep learning deep
learning is a machine learning technique which allows computers to do what
comes naturally to us learning by example deep learning is a key technology
behind self-driving cars it enables them to recognize lampposts and to tell the
difference between pedestrians and stop signs and it is used in so many other
ways in deep learning a computer model is trained using a large set of label
data and neural network architecture that contains many layers the term deep
refers to a number of layers in a neural network a traditional neural network
has about two to three layers but a deep learning neural network has over 150
and can have as many as thousands this is the reason why deep learning neural
networks can achieve such a high level of accuracy and can even beat humans at
recognizing things I'll go out well deeper into deep learning later on in
another Blog but for now let's move on to the five applications on computer
vision so the first really cool application is image classification image
classification is the process of identifying objects in images although it's
really easy for us humans it is an extremely difficult task for computer but
with deep learning computers are able to do that some really cool uses of image
classification is actually insecurities like identifying faces in CCTV footage
and even in real-time at airport another popular example of the use of image
classification is in self-driving cars because without image classification
self-driving cars would not be possible image classification allows
self-driving cars to identify where the road is and what all the obstacles are
so it's such a huge part of self-driving cars another example of image
classification is in the healthcare industry and how you can you're able to
identify tumors in scans of endoscopies and biopsies second application of
computer vision that I'm gonna be talking about its image colorization image
colorization does exactly what it says it basically adds color to black and
white.
images
however achieving the real color of a black and white image is extremely
difficult when we have a grayscale image we essentially only have one kind of
information about that image and that is its intensity so how do we determine
the color of each of those pixels in that image turns out each of the pixel in
a black and white image has 313 different color possibilities to choose from so
essentially this is the biggest problem that image colorization is trying to
solve so image colorization is used to color black-and-white images especially
old black and white images next up is 3d reconstruction for us humans
understanding how an object looks like in 3d when we're just looking at it in
2d is pretty easy however this is an extremely difficult task for computers
because it's hard for them to look at a 2d image and derive how it would look
like in 3d there are several ways in which 3d reconstruction can be I doubt one
of the most popular ways is to take several photographs of a singular object at
different angles and mesh them together to get a 3d image next up is image
synthesis image synthesis is an extremely broad genre essentially image
synthesis uses neural networks to generate images which are similar to existing
images in our data set but not exactly the same in 2017 by making use of image
synthesis researchers managed to generate photographs of human faces which
looks super realistic by using photos of celebrities in 2016 another paper
release showed how images could be generated from textual descriptions the
project use extremely detailed descriptions to generate images of birds and
flowers another popular example of image synthesis in the mainstream media is
the face app I'm sure many of you have tried it by now the face app uses image
synthesis to generate hyper realistic images of us when we are old the last
computer vision application that I'm going to be talking about is image style
transfer image child transfer is essentially taking two images and blending
them together to produce one it is used in so many ways especially in computer
graphics and media and many more style transfer is a technique which uses two
images the first one is the content image and the second one is the style
reference image it takes these two images and blends them together so that the
output image looks like the content image but painted in the style of the
reference image that's all for today's blog hope you guys found it interesting
and hope you guys learn something about computer vision. Thanks for read my
blog.
No comments: