Semantic segmentation is very crucial and self-driving cars and robotics because the models need to understand the context in the environment in which they’re operating Understanding this we have come for this semantic segmentation tutorial Now before we go ahead with the session elect to inform you that we have launched a completely free platform called Create Learning Academy where you have access to free courses such as Bay iCloud in Destin marketing.
It can check out the details in the description below the agenda harder The exception will be what is segmentation We would see what is the usual segmentation Uh what do we get out of it What other industry applications offered And Howard walls a post that we will do Ah one of the net Popular networks and the segmentation is probably we can say unit now I remember last time I showed you guys are due importance We remember them eminence today that we used the numbers for use I say it is a part off dimensionality reduction or the deep learning.
The same concept is going to come back to unit So if you can t collect that particular session on the court Probably I think we took 15 An image of good quick a session on that it is almost similar unit conceded work and not only unit there are many of these But for today’s session and our course united in the scope of in cedar Plus moon do one gay study onto this Ah After that the second part of it is uh be sure one more network called Mobile unit.
So I hope you people now are familiar with VG right Last time you didn t remember so we Jiji is a very heavy one Right everybody at least with that it cannot be used on a cell phone or a small thinkable device the harder it will not support What do we have You got a smaller lightweight version of that movie so we will see how we can get the same performance by using number ox multiplication and or whatever it star lord of layers in position Find about on DA after that What a thing they had roofs after the or get me on the check if anything’s ending after that if we get time so it’s gonna be a little heavy.
You get pain we will start with artists units Yeah So I’ll take your ways through what is our CNN And then I think I already showed you guys once But do it Redo it What is the past of CNN and what is Foster’s fostered out See how you gettin Bruising women So basically this is for object addiction This is also for object detection But in this, we do some something like a segment getting particle And there’s also holds up seen So now we’re moving from I would say classifications toe direction holding now coming to the first part of it What exactly is your understanding One unit What is happening Okay Uh So what are you doing here This well let us say we have an image Notice A This is going on and I won’t do Can I order this image into some kind of n corded structure which is smaller in size Yet it retains all the information over there So when I have such kind of problem what do we do The always bored and illusions.
So what I will do have big a larger image looks convolutions Max pulling and Commodore small And so just imagine the Haider’s did I mention I’m just reducing the dimension one by one All right so I love my deeper layers off revolution not evolution Pulling the evolution pulling city area I can say that the information that is stored here I have shrinking it down to this much This process is called including now what is the use of these things when we use First of all it could be used to store If in games you want to compress the Maidan and store it, yes you can do this Second thing is there were there could be sorted applications were a larger image or having ah you educate our obligation, not possibles, In that case, we will try to give this image.
Try to recreate something out of this back that was nothing but auto and borders People remember we got if you remember that numbers and all we had a political limit and the answer that we thought was a very big kind of downsample or against a very good example in Vegas even though from the looking at the image we can make out the numbers does not so bad So whenever you own dimension reduction on this thing could be lost but in place Case, we’re not looking for that application What We’re indoors I will now trying to three clear this image back Any idea How can I do that So let me first put the figure so that from there they know more near doing because I get back to the old is in damage So what I’m doing I’m doing compression-decompression.
So I consider this a Cordy This could be big Okay No my question to all of you is how we can do this How is it possible The convolution by people bought it We have listed The possibility how you can do this Let me unite’s so seeing I originally had a trying on it No what I did um when I encoded it I included a little smaller timers, Yeah Now my deal is bring that triangle back Such evidence it is almost similar to the original Something like this Yeah there isn’t What you do is the stock making you start So let us say that there are some pixels.
So you expand us because say that might be This is my current range The nearby itself over will be overseen So this could be one operation under double that One more person saying that Let me explain this under love off this one more something like this So you keep expanding are going in different directions Arterial in your Children Okay so this is your very busy content But if you want to look at from image point of the letter said this isn’t in cross an image Now where does this image how he’s a member-only numbers Nothing It’s after Including what What I’m gonna have to say I have my important They got to Boston It could be one Ah do pixel numbers four and six after including after going through multiple layers of the convention in the city and other books.
Now how do we get it back So whatever new I would like to be it’s something like a fired around it We’ll start doing the fighting on in this fighting What are our one day off-putting now When I add one little bad What’s gonna happen Getting others night So what if I say that the points near though this will get this money The points near the others will get this brother This is for four points near the sites since 111 Yeah They’re not gonna say yes There are some numbers around like this I will Keeping using 91 by one Yeah Now this looks convincing over here but are you able to find a problem.
There could be some problems If I do this there could be other John Boehner that might lie in this Correct Exactly So what I’m going to do because of my using the same pixels and replicating the same color Thanks Right Yes So because of this issue So actually it looks like an issue but yes we have got a solution here So why is the solution So now keep this in mind this type of concept and we’ll come back to this, Yeah So let me show It is problem There could be certain times there You don’t need to identify what is there in the image See in this image if you observe what are we can get out of If this image was passed to your YOLO your next week’s this week speak for printing you loving do you many object detection saying this is the house sky.