PatchGANReceptiveField

Enums · AiDotNet.Enums

The discriminator receptive-field sizes reported by pix2pix (Isola et al., 2017), section 6.1.2.

For Beginners

a PatchGAN discriminator does not judge a whole image as “real” or “fake”; it judges many overlapping square patches and averages the verdicts. This enum selects how big each judged patch is. Smaller patches police fine texture only; larger patches also police large-scale structure but need more parameters and are more prone to artifacts.

How It Works

The paper varies the patch size purely by discriminator DEPTH — every variant uses the same C64-C128-… pattern, just truncated or extended — so each member here maps to a layer count.

Fields

Field Summary
PatchGANReceptiveField Image286x286 286x286 “ImageGAN”: C64-C128-C256-C512-C512-C512.
PatchGANReceptiveField Patch16x16 16x16 PatchGAN: C64-C128.
PatchGANReceptiveField Patch70x70 70x70 PatchGAN: C64-C128-C256-C512.
PatchGANReceptiveField Pixel1x1 1x1 “PixelGAN”: C64-C128 with 1x1 spatial filters.
int value__