vae

→VAEModel

vae(pretrained: bool = False, overrides: object = {})

source edit

Construct a convolutional VAE with a single bottleneck latent.

Vanilla VAE topology following Kingma and Welling, 2014, configured with a 32x32 input, down_block_channels=(64, 128, 256) (three stride-2 encoder stages — 32 -> 16 -> 8 -> 4), and a 128-dimensional latent. Pass latent_dim as a tuple of integers to switch to the hierarchical topology, or call hvae directly to use the Sønderby 3-level stack default.

Pass kl_weight!=1.0 to obtain a $\beta$ -VAE (Higgins 2017).

Model Size

Parameters

pretrainedbool= False

Reserved for future weight registration; currently a no-op.

**overridesobject= {}

Optional VAEConfig field overrides (e.g. latent_dim=..., kl_weight=..., recon_loss="bce") forwarded into the underlying config.

Returns

VAEModel

Convolutional VAE trunk configured with the vanilla defaults and any overrides.

Notes

Reference: Kingma and Welling, "Auto-Encoding Variational Bayes", ICLR, 2014 (arXiv:1312.6114); $\beta$ -VAE in Higgins et al., "beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework", ICLR, 2017.

Reparameterisation trick:

z = \mu_\phi(x) + \sigma_\phi(x) \odot \epsilon, \qquad \epsilon \sim \mathcal{N}(0, \mathbf{I}).

Examples

>>> import lucid
>>> from lucid.models.generative.vae import vae
>>> model = vae().eval()
>>> x = lucid.randn((1, 3, 32, 32))
>>> out = model(x)
>>> out.sample.shape, out.latent.shape
((1, 3, 32, 32), (1, 128))

Used by 1

lucid.models.generative.vae

>>> import lucid >>> from lucid.models.generative.vae import vae >>> model = vae().eval() >>> x = lucid.randn((1, 3, 32, 32)) >>> out = model(x) >>> out.sample.shape, out.latent.shape ((1, 3, 32, 32), (1, 128))