Conv1DTranspose?

[hadaev8]

I want to make GAN for 1d object.
Whats best way to implement Conv1DTranspose?

Here write

Do not always use transpose operation for it will consume a lot of time.

So probably use Conv2DTranspose is a bad idea.

Here

@fbranchaud-charron-miovision

As Francois has explained multiple times already, a deconvolution layer is only a convolution layer with an upsampling. I don't think there is an official deconvolution layer. The result is the same.

But

@jwgu

The deconvolution is not equivalent as "convolution + upsampling".

So, how best to do this? Conv2DTranspose? UpSampling1D? Something else?

[OtterOrder]

Hi,

I have the exact same problem.
I'm not sure why there is no Conv1DTranspose in the first place since it proved pretty convenient for image generation for instance.
Also, I tried using UpSampling1D but unfortunately, it takes more and more time the more I add layers until it finally crashes.

@hadaev8 Did you try with UpSampling1D? Did it work for you?

[hadaev8]

@OtterOrder
not yet tried

[MinnML]

@OtterOrder and @hadaev8 , I have tried both UpSampling1D+Conv1D and also Conv2DTranspose with a 1D kernel on 1D data. They were pretty much the same in speed and accuracy.

[MatthewAwesome]

This feature has been implemented in tf.keras, (tf v2.2.0), check it out!