ncsn_celeba_gen

→NCSNForImageGeneration

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

source edit

Construct an NCSN CelebA 64x64 model with DSM loss and .generate().

Same trunk as ncsn_celeba (CelebA 64x64 setup, $L = 500$ noise levels), wrapped with the denoising score-matching loss and annealed Langevin sampling.

Model Size

Parameters

pretrainedbool= False

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

**overridesobject= {}

Optional NCSNConfig field overrides forwarded into the underlying config.

Returns

NCSNForImageGeneration

CelebA 64x64 NCSN wrapped with the DSM loss head and Langevin sampler.

Notes

Reference: Song and Ermon, "Improved Techniques for Training Score-Based Generative Models" (NCSNv2), NeurIPS, 2020 (arXiv:2006.09011), Appendix C.

Examples

>>> import lucid
>>> from lucid.models.generative.ncsn import ncsn_celeba_gen
>>> model = ncsn_celeba_gen().eval()
>>> x = lucid.randn((1, 3, 64, 64))
>>> out = model(x)
>>> out.sample.shape   # raw score field
(1, 3, 64, 64)

Used by 1

lucid.models.generative.ncsn

>>> import lucid >>> from lucid.models.generative.ncsn import ncsn_celeba_gen >>> model = ncsn_celeba_gen().eval() >>> x = lucid.randn((1, 3, 64, 64)) >>> out = model(x) >>> out.sample.shape # raw score field (1, 3, 64, 64)