Implementing Adversarial Latent Autoencoders [CVPR2020]

[frostbyte012]

Short Description

Autoencoder networks are unsupervised approaches aiming at combining generative and representational properties by learning simultaneously an encoder-generator map.

Papers

https://arxiv.org/pdf/2004.04467.pdf

Existing Implementations

https://github.com/podgorskiy/ALAE

Other Information

One implementation will ensure creating 1024x1024
The resolution can also produce face reconstructions and manipulations based on real images.

[frostbyte012]

@innat, @LukeWood @bhack I was looking at some awesome examples and came by it, can you all confirm the feasibility of this implementation to be added here in Keras CV ? A Py-Torch implementation has been provided above.

[LukeWood]

I don't think that we have anything on the roadmap for offering algorithms like this. I'd guess we are more likely to offer SimCLR or something like that than this in the near future.

[innat]

@frostbyte012 It might possible to add if keras-cv start supporting GAN based model in the future, LIKE.

@LukeWood I think TF-Similarity already offers SimCLR and mutual algorithms.

[LukeWood]

I think for now we can close this. These sorts of applications are at least a year off.

[frostbyte012]

@LukeWood Sure Sir! But I'd love to see it implement in kerasCV soon someday!