def build_discriminator_trainer(self):
"""
Build a Keras model for training image and mask discriminators.
"""
# Mask Discriminator
D_Mask = Discriminator(self.conf.input_shape,
output='2D',
downsample_blocks=3,
name='D_M')
D_Mask.build()
self.MaskDiscriminator = D_Mask.model
real_M = Input(self.conf.input_shape)
fake_M = Input(self.conf.input_shape)
dis_real_M = self.MaskDiscriminator(real_M)
dis_fake_M = self.MaskDiscriminator(fake_M)
D_Image = Discriminator(self.conf.input_shape,
output='2D',
downsample_blocks=3,
name='D_X')
D_Image.build()
self.ImageDiscriminator = D_Image.model
real_X = Input(self.conf.input_shape)
fake_X = Input(self.conf.input_shape)
dis_real_X = self.ImageDiscriminator(real_X)
dis_fake_X = self.ImageDiscriminator(fake_X)
self.D_model = Model(
inputs=[real_M, fake_M, real_X, fake_X],
outputs=[dis_real_M, dis_fake_M, dis_real_X, dis_fake_X])
self.D_model.compile(Adam(lr=0.0001, beta_1=0.5), loss='mse')
log.info('Discriminators Trainer')
self.D_model.summary(print_fn=log.info)
def build_mask_discriminator(self):
# Build a discriminator for masks.
D = Discriminator(self.conf.d_mask_params)
D.build()
log.info('Mask Discriminator D_M')
D.model.summary(print_fn=log.info)
self.D_Mask = D.model
real_M = Input(self.conf.d_mask_params.input_shape)
fake_M = Input(self.conf.d_mask_params.input_shape)
real = self.D_Mask(real_M)
fake = self.D_Mask(fake_M)
self.D_Mask_trainer = Model([real_M, fake_M], [real, fake],
name='D_Mask_trainer')
self.D_Mask_trainer.compile(Adam(lr=self.conf.d_mask_params.lr),
loss='mse')
self.D_Mask_trainer.summary(print_fn=log.info)
def build_image_discriminator2(self):
"""
Build a discriminator for images
"""
params2 = self.conf.d_image_params
params2['name'] = 'D_Image2'
D = Discriminator(params2)
D.build()
log.info('Image Discriminator D_I2')
D.model.summary(print_fn=log.info)
self.D_Image2 = D.model
real_x = Input(self.conf.d_image_params.input_shape)
fake_x = Input(self.conf.d_image_params.input_shape)
real = self.D_Image2(real_x)
fake = self.D_Image2(fake_x)
self.D_Image2_trainer = Model([real_x, fake_x], [real, fake],
name='D_Image2_trainer')
self.D_Image2_trainer.compile(Adam(lr=self.conf.d_image_params.lr),
loss='mse')
self.D_Image2_trainer.summary(print_fn=log.info)