autoencoders.rst

Autoencoders

This section houses autoencoders and variational autoencoders.

---

Basic AE

This is the simplest autoencoder. You can use it like so

from pl_bolts.models.autoencoders import AE

model = AE()
trainer = Trainer()
trainer.fit(model)

You can override any part of this AE to build your own variation.

from pl_bolts.models.autoencoders import AE

class MyAEFlavor(AE):

    def init_encoder(self, hidden_dim, latent_dim, input_width, input_height):
        encoder = YourSuperFancyEncoder(...)
        return encoder

pl_bolts.models.autoencoders.AE

---

Variational Autoencoders

Basic VAE

Use the VAE like so.

from pl_bolts.models.autoencoders import VAE

model = VAE()
trainer = Trainer()
trainer.fit(model)

You can override any part of this VAE to build your own variation.

from pl_bolts.models.autoencoders import VAE

class MyVAEFlavor(VAE):

    def get_posterior(self, mu, std):
        # do something other than the default
        # P = self.get_distribution(self.prior, loc=torch.zeros_like(mu), scale=torch.ones_like(std))

        return P

pl_bolts.models.autoencoders.VAE