Depthwise Separable Convolution - A FASTER CONVOLUTION!

[@masoudmasoumimoghaddam3832]

I love the videos you make. The good thing is that you explain the concept right to the point and don't waste time which means you are dominant on the subject. I really hope you don't lose the motivation to make such tutorials. because there are enthusiasts like me and my colleagues that literally waiting for your future videos. So please keep making videos

[@sarthaknarayan2159]

Your channel is underrated and is pure gold

[@IndiaNirvana]

Very crisp explanation loved it.

[@VikasSingh8]

love what you are doing, your recent videos were really helpful to me, keep up the good work, keep exploring and uploading videos 👍

[@CodeEmporium]

Thanks! Glad you like the videos!

[@王大哥-e1c]

thanks for your high-quality videos which really help me a lot

[@GunnuBhaiya]

This is a wonderful tutorial which deserves (and in the future will get) way more views

[@CodeEmporium]

Rahul Gore Hoping the same. Thanks!

[@ParniaSh]

Well explained, beautifully demonstrated. Thanks!

[@starcraftpain]

Finally understood MobileNets and DSCs. Thanks the for clear video!

[@deepaksingh5607]

You explained it in the best way.

[@7688abinash]

That put down so simply. Just loved it :) Thanks a lot

[@CodeEmporium]

Abinash Ankit Raut glad you liked it! Thanks!

[@esrabetulelhusseini2330]

provide a clear understanding to me .so glad,thank you

[@myhofficiel4612]

well explained , you made it look really easy !

[@ShahriarMohammadShoyeb]

Brilliant explanation, described in a very understandable way.

[@CodeEmporium]

Shahriar Mohammad Shoyeb thanks! Glad you liked it !

[@vamsiKRISHNA-io1yi]

Simply Brilliant thank you for much for a detailed information about Xception

[@maxsch.2367]

absolute banger! well done

[@张程-f6d]

really helpful for me to understand the depthwise separable convolution! Thank you!

[@thecurious926]

omg this came out 4 years ago? I am living under a rock

[@mamuncse068]

Excellent video,easy and well explain of Depth wise Separable Convolution.Really grateful to you.

[@keithchua1723]

Imbeccable explanations as always!

[@Vinay1272]

Thanks a lot for this! Very helpful.

[@senli2229]

Great help for understanding DepthWise Seqarable Convolution!!!

[@jeamsdere9636]

This video is real helpful. thank you

[@sami-h5y]

That was a very lucid explanation, thanks.

[@CodeEmporium]

Glad you found it usefule Sangeet

[@Kumar731995]

Awesome video, subscribed! Had a really good understanding of what depth wise separable convolution is at the end of the video.

[@huythai6210]

It is so useful and clear

[@knowhowww]

Okay, now I get it.... Thanks!

[@yungwoo729]

Perfect explanation. I appreciate it. Thank you!

[@dufrewu7437]

very helpful video, thanks

[@AmartyaMandal7]

Amazing Explanation!

[@zhuotunzhu8660]

Very clear, make it easy to understand! Thanks!

[@CodeEmporium]

Zhuotun Zhu anytime! Thanks for watching

[@Frostbyte-Game-Studio]

this is fantastic explaination

[@CodeEmporium]

Thank you so much!!

[@RafiqulIslam-je4zy]

Many many thanks.

[@roonyyu5374]

thank you very much.

[@duongkstn]

incredible !

[@harutyunyansaten]

thank you, understood

[@Ganitadava]

Super explanation

[@vinitakumari5913]

Explained it so simply. Thanx

[@CodeEmporium]

No worries. Glad it helps!

[@austinmw89]

Best explanation I've found on this, thanks

[@CodeEmporium]

As long as it helps :)

[@virajwadhwa6782]

Is standard convolution here and depth-wise separable convolution functionally equivalent? That is, they will both give the same outputs for a certain input? It is just that, depth-wise separable convolution saves on computations, but is otherwise functionally the same right?

[@felippewick]

Great video. Helped a lot!

[@UniversalRankingOfficial]

Can you make a video on Resnet Architecture for beginners?

[@PalashKarmore]

Thank you. You saved me a lot of time.

[@CodeEmporium]

It's what it do. Thanks for watching :)

[@bearflamewind]

Thank you so much for making such a nice video that is so easy to understand.

[@CodeEmporium]

GUO GUANHUA For Sure! I'm glad you understood it :)

[@win-n6k]

great video,looking forward more

[@maheshwaranumapathy]

Great video, reading the reference paper is going to be much easier now

[@roberttlange8607]

Great explanation! Thank you very much!

[@willz3222]

This is excellent

[@gaussian3750]

Thanks for explanation

[@mayankchaurasia4483]

Awesome explanation . Loved it.

[@CodeEmporium]

Mayank Chaurasia So glad you loved it :)

[@sahibsingh1563]

Awesome explanation

[@vaibhavsingh1049]

This was great.

[@Maciek17PL]

What would pointwise convolution look like in a 1d separeble convolution???

[@reactorscience]

Amazing video sir.

[@RishabhGoyal]

Very clear explanation.. Thanks a lot.

[@CodeEmporium]

Welcome! Glad you got some use out of it

[@RishabhGoyal]

CodeEmporium Yeah.. I was reading W-Net where they have used it..

[@sergeyi2518]

Is it correct that arbitrary standard convolition cannot be exposed as depthwise convolution (except some special cases)? Depthwise convolution is just another type of convolution, right?

[@srinathkumar1452]

Nice video! I look forward to future videos on object detection and semantic segmentation.

[@sounakbhowmik2841]

thank you, it was of great help !!

[@wuxb09]

Good Explanation! Thanks

[@busy_beaver]

Thanks!

[@Lucas7Martins]

Loved it!

[@lonewolf2547]

Awesome video dude

[@harv609]

Amazing .. explained so clearly !! Thank you

[@CodeEmporium]

Harsha Vardhana anytime! Glad you liked it!

[@meyouanddata9338]

amazing content. thanks alot :)

[@MasayoMusic]

Thank you for this. What are you using for animations?

[@ducpham9991]

very clear!

[@yx9873]

Well. I can't understand why the input size of the second phase is still M. Is that a typo?

[@zhengxiangyan3654]

excellent！very nice video

[@gopsda]

Great! Neatly put. Thanks for the video. One thought -- we can add one parameter lambda as multiplication factor in the combination step, and treat as a trainable parameter which increases total trainable parameters by 1 but may help converge the solution faster, I guess. Depthwise sep conv = Depthwise conv + lambda * Pointwise conv.

[@strongsyedaa7378]

Where to use depthwise separable convolution? How do we come to know to where to use it? 🤔

[@gopsda]

@@strongsyedaa7378 Wherever you want to reduce number of trainable parameters. Most of the networks are defined with this depthwise conv.

[@artsyfadwa]

Nice video. Thanks.

[@bhuvneshkumar1970]

2:00 shouldn't it be (Dk^3 ) * M? As matrix multiplication of size (n x m) . (m x p), no. of multiplication are n x m x p.

[@sourishsarkar5281]

Why are the output number of features always an integral multiple of the number of input channels?

[@ah-rdk]

Thanks.

[@nexushotaru]

Thank you for explanation, but please, use more intuitive designations (like H for height and W for width)

[@abhishekchaudhary6975]

Thanks

[@66tuananh88]

Do you have a python code 3d depthwise separable convolution?

[@melihaslan9509]

very nice!

[@digvijayyadav3633]

worth the time!!

[@jo-of-joey]

omg. you just saved the day!

[@CodeEmporium]

You can always count on your friendly neighborhood data scientist..

[@jo-of-joey]

can you do a video on Binarized Neural Networks?

[@rabhinav]

Hey really helpful Thank You. Can you also make a video on Winograd Convolution?

[@strongsyedaa7378]

Where to use depthwise separable convolution?

[@joruPT]

This video was very helpful, thank you :)

[@CodeEmporium]

Welcome. Glad it was useful!

[@judyhwang9342]

excellent

[@CodeEmporium]

Thanks!

[@kartikpodugu]

easy to understand. i suggest to add animations for better understanding if possible. thanks

[@mohitpilkhan7003]

"immediately" hahaha. Thanks bro. SUbscribed

[@ranam]

Ok genius iam also approaching the problem same way like you I don't use matheMatical way my question is so simple because LTI depends on convulution here's my question below Convulution is nothing but stacks and scales the input that's why the input to an amplifier is stacked and scaled or amplified but in filter design it attenuate frequency so I don't know how it regret certain frequency by stacking and scaling the input if possible some one explain to me

[@varchitalalwani3802]

very helpful, thanks

[@CodeEmporium]

Glad it was helpful. Thanks for watching!

[@willemprins4564]

How does this do with Res and Densenets?

[@harshnigm8759]

In the Xception research paper they actually used skip connections and dense layers , skip connection were reported to have given a major boost to the final accuracy.

[@aq555]

good

[@harshadj13]

Sakkath video!

[@pawansj7881]

Good1

[@rahuldeora5815]

Hey, I am making a video using some of your animations. Hope its cool!? It's on MobileNets

[@CodeEmporium]

bluesky314 Absolutely. Just list this video in your references. Send a link to your video here when you're done. I'd like to see it :)

[@rahuldeora5815]

Thanks! Here it is: kzbin.info/www/bejne/fnWcd6Ggi9yIbsk Would love your feedback

[@rahuldeora5815]

hey

[@mathematicalninja2756]

Tjis is like mapreduce

[@DarkLordAli95]

First, thank you for making this helpful video. Second, why can't comp sci people agree on one notation for anything at all?! It's like for every video I watch I gotta learn a new set of notations... BOY. And why is F the input and not the filters? that's just straight up confusing man. humans really can't agree on anything.

[@Jonas-qz2gb]

Thank you so much for this amazing explanation!

[@santhoshkolloju]

Very helpful Thank you

[@CodeEmporium]

Thanks! Glad it was of use.