realnvp_imagenet32(pretrained: bool = False, overrides: object = {})Construct the RealNVP flow for the Imagenet 32 x 32 setup.
Paper-faithful configuration from Dinh et al., 2016 §4.1: four scales
(so the recursion ends on a 4 x 4 map), coupling networks of 4
residual blocks with 32 feature maps at the first scale, doubling per
scale.
Model Size
Parameters
pretrainedbool= FalseReserved for future weight registration; currently a no-op.
**overridesobject= {}Optional
RealNVPConfig field overrides forwarded into
the underlying config.Returns
RealNVPModelBare bijection configured with the Imagenet 32 x 32 setup.
Notes
Reference: Dinh, Sohl-Dickstein, and Bengio, "Density Estimation Using Real NVP", ICLR, 2017 (arXiv:1605.08803), §4.1 and Table 1. Reported validation result: 4.28 bits/dim (4.26 train).
Examples
>>> from lucid.models.generative.realnvp import realnvp_imagenet32
>>> model = realnvp_imagenet32().eval()
>>> model.config.num_scales
4