Transposed Convolution

[@EgonHoltz]

After a lot of texts trying to understand how it works, this video just saved me. Thank you for share the knowledge!

[@tejaszodge4211]

The wittiest ML guy ever XD. People are in red, maybe because they are probably communists XD @2.04

[@leopoldodellaporta3831]

Nice to see well made explanations for sub-elements of DL! Clear, short and to the point, thanks!

[@TheJysN]

So Transposed convolution is deconvolution, got it, thank you!

[@ambujmittal6824]

lol

[@jorisvanlienen8891]

Please never say that again

[@chrisdahms9682]

🙃

[@divyab2829]

Thanks for the clear explanation with an example

[@AbdullahOlcay-wh4or]

That's a very good explanation of deconvolution. Many thanks.

[@notogp]

That is NOT deconvolution 😅

[@AbdullahOlcay-wh4or]

:D@@notogp

[@navid2368]

So transposed convolutions are used to detect communists, got it!
Thanks for the explanation :)

[@yashthakre]

02:02 what i wasnt expecting that

[@coffeenomilk6614]

Thank you, this was a very good explanation!

[@rosenman22]

now that's what i call hand's on explanation! well done!

[@sri_dharsh_ni3749]

Finally understood this topic
Thank you 😌😌

[@swarajnanda7874]

Thanks a lot! This helps me understand the UNet's upscaling step a lot better!

[@nkapila6]

Thanks for this, very helpful!

[@usamazahid1]

very good illustrations and to the point explanation

[@anirudhrvs4849]

Simplest and the best explanation. Kindly make more videos.

[@thilinawee]

Thanks Joris. Simple and clean

[@alalehrazmjoo2850]

Thanks for the clear explanation!

[@alanblanchet6814]

2:02 "They are probably communists" 🤣

[@Democracy_Manifest]

Great video, thanks.

[@jennys3960]

This is the definition of convolution. So what is the difference between Transposed convolution and regular convolution?

[@simonhorna5081]

no, Its not. In convolution you overlap the kernel on the input and multiply the overlaping values

[@arminatayeba9613]

In the given example, how have we already known that the output matrix is 3*3 and why not 4*4?

[@mahletalem]

Thanks! This is clear and easy to understand.

[@dingusagar]

the last example by calculation was the one which made it clear. thanks.

[@user-nx8ux5ls7q]

Sorry, I am new to this. First time hearing of Transposed Convolution. At 4:17 only one column from each matrix are overlapping. Is it because the stride is 1? If stride is 2, there won't be any overlapping? Thank you. Please be gentle with the reply guys I am a newbie trying to learn.

[@НиколайМакаров-ж1щ]

You are totally correct, there is overlapping because of stride 1, with stride=2, it won't be any overlapping

[@user-nx8ux5ls7q]

@@НиколайМакаров-ж1щ Thank you!

[@morgengabe1]

does anybody have good advice for setting these parameters in general? is there a special reason this is chosen?

[@V.aravind]

thanks bro, was really helpful

[@RKYT0]

ok whats the difference to deconvolution?

[@abdolvakilfazli2488]

okay, that was clear and thank you

[@vikashranjan5210]

viery nicely explained... Thank you so much , Joris !

[@tingnews7273]

so the picture took in iKEA?

[@morgengabe1]

for the example, he wants to transform a 2x2 image to a 4x4. he uses a 3x3. how does he choose the size 3x3? could you choose any size? or is it he only square greater than 2x2 and smaller than 4x4?

[@aaomms7986]

Thank!! nice explanation

[@sujathaontheweb3740]

Understoooood! So happy! 😀😀😀 Thank youuuu! Even a bozo like me can understand when taught like this!!! Thank youuuu!

[@eliethesaiyan]

does conv2DTranspose simulate feature x kernel instead of kernel x feature?

[@Andionschannel]

Great explanation!

[@jacob-un1ov]

Thank you so much!

[@youtubecommenter5122]

Those people in red are probably communistsXD

[@mhmdnojim9479]

amazing ,,,got it

[@BlueQualityRhythm]

Great explanation. You can improve on your audio Quality however.

[@talha_anwar]

How you produce an output matrix of 3x3 instead of 4x4?

[@morgengabe1]

did you ever find out the way to choose the intermediate matrix?

[@thegt]

"They're probably communists" LMAO

[@vibhudalal901]

2:02 😂

[@CppExpedition]

Thx J!

[@tribuiduonguc7788]

thanks a lot

[@florentb8578]

Thank you

[@luisfable]

nice, thanks

[@mehrdadkazemi3969]

ty

[@mager8460]

2:01 Hahaha!!!

[@norman9174]

thanks

[@breakdancerQ]

Why Israel?

[@jorisvanlienen8891]

Because it is a holy country

[@muhammadsarimmehdi]

you should do another example with a bigger kernel and non-zero padding. Your method is not clear with the example you provided

[@morgengabe1]

did you ever figure out how to pick a good kernel size?

[@tusharverma3788]

probably communist🤣

[@maxnord9356]

Bla bla bla 😂

[@Banefane]

Thanks for the clear explanation!