|
import tensorflow as tf
import hyperchamber as hc
import numpy as np
from hypergan.losses.standard_loss import StandardLoss
from hypergan.losses.least_squares_loss import LeastSquaresLoss
from hypergan.losses.wasserstein_loss import WassersteinLoss
from hypergan.losses.base_loss import BaseLoss
class LambGanLoss(BaseLoss):
def required(self):
return "label_smooth".split()
def _create(self, d_real, d_fake):
config = self.config
alpha = config.alpha
beta = config.beta
wgan_loss_d, wgan_loss_g = WassersteinLoss._create(self, d_real, d_fake)
lsgan_loss_d, lsgan_loss_g = LeastSquaresLoss._create(self, d_real, d_fake)
standard_loss_d, standard_loss_g = StandardLoss._create(self, d_real, d_fake)
total = min(alpha + beta,1)
d_loss = wgan_loss_d*alpha + lsgan_loss_d*beta + (1-total)*standard_loss_d
g_loss = wgan_loss_g*alpha + lsgan_loss_g*beta + (1-total)*standard_loss_g
return [d_loss, g_loss]
|
