Tutorial 24- Max Pooling Layer In CNN

Please join as a member in my channel to get additional benefits like materials in Data Science, live streaming for Members and many more
/ @krishnaik06 In this video we will understand about the max pooling layer in CNN
Complete Deep Learning playlist:
• Tutorial 1- Introducti...
Connect with me here:
Twitter: / krishnaik06
Facebook: / krishnaik06
instagram: / krishnaik06
Subscribe my unboxing Channel
/ @krishnaikhindi
🙏🙏🙏🙏🙏🙏🙏🙏
YOU JUST NEED TO DO
3 THINGS to support my channel
LIKE
SHARE
&
SUBSCRIBE
TO MY CZcams CHANNEL

[@user-gx9hk8gt3k]

I don't know who came up with this Max Pooling but he must be a genius. Thank you for the video!

[@sandipansarkar9211]

Superb video.Read a lot and saw videos of maxpooling but this one cleared all my doubts.Thanks Krish. Keep it up.Cheers.

[@mukund198526]

Really so simply explained and now see the difference how a upgrad professor explained the same concept -
Max pooling: If any one of the patches says something strongly about the presence of a certain feature, then the pooling layer counts that feature as 'detected'.
Average pooling: If one patch says something very firmly but the other ones disagree, the pooling layer takes the average to find out.

[@sanjivchindarkar]

true that xd

[@Gester2000]

This guy is a legend of the game I was watching 7 hours of deep learning video in which CNN WAS 1 HOUR AND my doubts were still not cleared this guy did it in few minutes I am highly impressed by your skills Sir

[@koushikshomchoudhury9108]

Awesome explanation & thank you.Highly inefficient channels like *DONT WANT TO TAKE THE NAME* takes thousands of rupees and teaches with about 10% proficiency as you do. This will take me to a step closer to my paper. :)

[@Tales.of.Irshad]

whenever i have doubts... i visithere...go back with good knowledge

[@vishaljhaveri7565]

Thank you, Krish Sir. Nice tutorial on max pooling.

[@MsGeetha123]

Excellent Sir.. thank u so much

[@arjunsaravanan4855]

Great explanation!

[@meghachoudhary3394]

Hi Krish, please continue your deep learning series.

[@harendrakumar7647]

Hey Krish, Can you please explain about the strides and How to set up the values for strides in tensorflow ? Thanks

[@aidataverse]

great explanation

[@louerleseigneur4532]

Thanks Krish

[@michaelscience2481]

Awesome explanation

[@EssDubz]

In no existing framework anyway does max pooling round up the output dimension size. If stride takes you off the edge, you don't include it. The output dimension for a 3x3 image, with a kernel size of 2, and stride of 2, is 1x1

[@vallurirajesh]

Thank you.

[@ravibhushandixit3500]

Thanks

[@varuntrivedi637]

Superb

[@srinathraghavan2866]

Thank you sir

[@Pr3kashSingh]

Thanks sir .

[@harshays2873]

sir we need remaining theory part and coding part of CNN ,please...

[@loaialamro9699]

thank you so much for this explanation, can you please provide the formula of the Max-pooling

[@alipaloda9571]

sir what if we chose filter size bigger than image size? is that filter size is hyper parameter if not then how to choose filter size?

[@smurtiranjansahu5657]

Sir pls make video about the CNN project...

[@kunalsutar3946]

excellent!

[@memesymubrra2472]

ty

[@badiyabhargav8597]

Sir, Can any one plz clarify my doubt that
do we apply activation for max pooling ?
or
we apply activation fun before pooling method ?

[@StevenSmith68828]

This is going to sound dumb but in a 2x2 how would a 1d max pooling work with size 1 would that just return the same thing or the highest number I have been searching and have not found a good answer

[@alirizvi3506]

KRISH>ANDREW NG LOVE FROM PAKISTAN!

[@aditisrivastava7079]

Am confused..... We do padding so that dimension will not reduce then we do max pooling that reduces the dimension....... Though I understood the very purpose of max pooling but this dimension reduction process making me confused

[@yerriswamyv1910]

I think, padding helps to detect the all edges of items n prefers in 1st layer and as we go ahead into further convolution layers have to approximate the process of identification where max pooling will helps.

[@aakankshajaiswal1809]

You apply padding in the convolution layers to prevent the loss of valuable information at the edges. As we move deeper into the hidden layers, after the extraction of important featurs, we need to reduce the dimensionality because further propagation of these volumes is not very reasonable. Also, once we have detected some featues already, there comes a time when we need to pick the brightest pixel from all the divided regions to get a clearer view of the entire image, like what has been detected in overall input image. That is why we pick "High pixel intensities" as they represent their neighbourhood.

[@anilpise4647]

Sir when you uploading the next videos?

[@brown_bread]

There are no trainable param in pooling layers. how NN will update pooling filter?

[@nashas6778]

How to apply max pooling on any image data set?

[@prateeksomani5803]

What will be the stride if we use 3×3 filter ?

[@shaz-z506]

Hi Krish,
Please let me know, in what scenario we should use average pooling over Max pooling.

[@girishreddyedula2667]

Shehzada use only max pooling

[@rithwikchhugani8117]

Hi krish. I am confused about one thing. Once we have applied the filter on the image, does it pass through the activation function and then go to the maxpooling layer or the activation function is applied twice ?

[@MrAyandebnath]

After convolution, activation function is used and after that max pooling is used. Activation function is not used twice.

[@shanmukhavarma3361]

How to implement this

[@vermaanky11]

As per your previous video, you informed that if padding layer is added then the formula is n-2p-f+1. Hence if we apply the same here with P=1, then we should get 1X1 matrix rather than 3X3. Correct me if I am wrong.

[@aritraray2501]

it's n - f + 2p + 1...that shd actually give 5

[@omingole7304]

@@aritraray2501 Ya, it's (n+2p-f)/s+1 , right?

[@ahlamsaeed498]

@@aritraray2501 yeah output is 5

[@mdmynuddin1888]

Please provide the research paper link

[@rahuldey6369]

What if I take stride=1, what will be the problem?

[@aditisrivastava7079]

Please provide the link for the research paper you were talking about

[@sargun_narula]

Just do a google search for "yann lecun cnn paper" or go to yann.lecun.com here you will find all his papers and publications

[@ashutoshsrivastava416]

In the video I/L is 4x4 and filter is 2x2 , padding is 1 , stride is 1 and the output is 3x3 but in the previous video the formula told by him is n+2p-f+1 if we get output as 5 how come can anybody explain me this ...

[@arkeshashah4650]

Sir, can you make the video on how cnn work with text classification

[@HarshSharma-mp7zf]

aisa na boliye

[@aakankshajaiswal1809]

I think that the filter dimension is a hyperparameter that is fixed and cannot be updated during backpropagation. Still not sure, correct me if I'm wrong.

[@krishnaik06]

Filter dimension will not be changed only the filter inner values will be changed

[@misslollipop1571]

Hero

[@ashwinimandani2829]

Sir what is Stride here?

[@AbhaySingh-fp6ew]

At 5:50, are you saying that max pooling layer is also learned during the training process? if so, then that seems wrong

[@sohinimitra7559]

+1, pooling layer has no parameters to learn. There is no update during gradient descent for pooling.

[@robinkumar9021]

Is it really will jump like that ?

[@cr7themachine940]

can you please provide a rp of cnn

[@yashgupta417]

sir, I think you forgot to consider padding in determining output

[@jagdishjazzy]

Yes although he mentioned about padding but not considered for this convolution layer

[@RanjitSingh-rq1qx]

Sir if we apply the padding equal to 1 then we will get 4*4 metric output. Not a 3*3 metric output. I learnt this thing in your previous video. But u r saying now it will return 3*3 metric. how is it possible sir ?

[@mansishah3957]

Same doubt

[@robinkumar9021]

Please provide the research paper

[@akshatsingh6036]

sir what is STRIDE ??

[@pankajyadav-en7tb]

Hi sir, really awesome explanation, but just one question did someone hit you before creating this video? I can see injury marks on your face.

[@224_harsh2]

:)

[@RAZZKIRAN]

please provide research paper

[@bhanuPrakash-yo5wd]

Sir complete the cnn part with one project of open CV image segmentation

[@krishnaik06]

Don't worry it will come in the advanced cnn section

[@bhanuPrakash-yo5wd]

@@krishnaik06 Thank you sir,hope you done soon ,I was in final this was the project I am working for my resume

[@hassannaqvi3451]

3:40 wouldn't we take the padding?

[@chanmad]

The matrix he's referring to is most likely after the filter has been applied. Padding is on the original image matrix, on which filter is applied.

[@lakshyasingh9408]

​@@chanmad ​ Still convoluted output will be 5x5. After padding input is 6x6 so, i=6,f=2 then ((6-2)/1)+1=5 .

[@tanushsingh2782]

Isn't the formula supposed to be (n +2p - f)/s + 1 ?

[@mansishah3957]

Same doubt

[@sagar140]

can anyone tell me what max pooling of size 1x1 do?

[@aakankshajaiswal1809]

Nothing absolutely.

[@sonamtenzin3457]

The input to a pooling layer has a width, height and depth of 224x224x3 respectively. The pooling layer has the following properties:
Kernel shape: 2x2
Stride: 2
PLEASE HELP ME

[@anandamoymukherjee6164]

Here stride =2

[@abhishekpurohit3442]

CNN is a bit confusing than ANN...

[@rajarshibasak559]

not clear..do we apply max pooling in output!!i mean max pooling is not clear.first video which is this much unclear to me.

[@IAmTheMainCharacter]

Sir please first continue with ml in 90 days, after that we can learn deep learning.

[@pranjalgupta9427]

3:40

[@mauricioalfaro9406]

Hello. Could you please speak more slowly?

[@Gester2000]

This guy is a legend of the game I was watching 7 hours of deep learning video in which CNN WAS 1 HOUR AND my doubts were still not cleared this guy did it in few minutes I am highly impressed by your skills Sir

[@abhinavkaushik6817]

great explanation