realnvp_lsun(pretrained: bool = False, overrides: object = {})Construct the RealNVP flow for the LSUN 64 x 64 setup.
Paper-faithful configuration from Dinh et al., 2016 §4.1 — the same
five-scale, 2-residual-block architecture used for every 64 x 64
dataset. One factory covers the three reported LSUN categories
(bedroom, tower, church outdoor); they differ only in training data.
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 LSUN 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 results: 2.72 (bedroom), 2.81 (tower) and
3.08 (church outdoor) bits/dim. Images are downsampled so the
smallest side is 96 px, then randomly cropped to 64 x 64.
Examples
>>> from lucid.models.generative.realnvp import realnvp_lsun
>>> model = realnvp_lsun().eval()
>>> model.config.residual_blocks
2