Image as input for VAE?

vision
1

Hello mate’s,
I have a question on VAE.
I’m working on a project for image to image translation (colourization) and I wanna use the VAE to benefit from the two losses (KLD and MSE).
Can I use VAE for a problem as colourization? Because all example’s are using noise input.
Thank you,

2

I suggest using GANs or plain CNNs. Due to the instinct of VEA, the latent vector must be sampled from a Gaussian distribution N(0,1), otherwise you would not be able to train your model.