def __init__(self, options): # line 91 in train.py calls model.GAN(model_options). # These model_options are the assigned to self.options in this class. self.options = options # Creating batch normalization layers (from line 41-44 for generator and 46-49 for discriminator): # "batch normalization reduces the internal covariance shift" # It makes the learning of layers in the network more independent of each other. # The objective of batch norm layer is to make input to the activation layer, unit Gaussian. # So that neuron does not get saturate in case of sigmoid and tanh. # It helps in the following: # Fast convergence of network. # Allows you to be care free about weight initialization. # Works as regularization. # Batch norm layers for generator self.g_bn0 = ops.batch_norm(name='g_bn0') self.g_bn1 = ops.batch_norm(name='g_bn1') self.g_bn2 = ops.batch_norm(name='g_bn2') self.g_bn3 = ops.batch_norm(name='g_bn3') # Batch norm layer for descriminator self.d_bn1 = ops.batch_norm(name='d_bn1') self.d_bn2 = ops.batch_norm(name='d_bn2') self.d_bn3 = ops.batch_norm(name='d_bn3') self.d_bn4 = ops.batch_norm(name='d_bn4')
def __init__(self, options): self.options = options self.g_bn0 = ops.batch_norm(name='g_bn0') self.g_bn1 = ops.batch_norm(name='g_bn1') self.g_bn2 = ops.batch_norm(name='g_bn2') self.g_bn3 = ops.batch_norm(name='g_bn3') self.g_bn4 = ops.batch_norm(name='g_bn4') self.d_bn1 = ops.batch_norm(name='d_bn1') self.d_bn2 = ops.batch_norm(name='d_bn2') self.d_bn3 = ops.batch_norm(name='d_bn3') self.d_bn4 = ops.batch_norm(name='d_bn4')
def __init__(self, options): self.options = options self.g_bn0 = ops.batch_norm(name='g_bn0') self.g_bn1 = ops.batch_norm(name='g_bn1') self.g_bn2 = ops.batch_norm(name='g_bn2') self.g_bn3 = ops.batch_norm(name='g_bn3') self.d_bn1 = ops.batch_norm(name='d_bn1') self.d_bn2 = ops.batch_norm(name='d_bn2') self.d_bn3 = ops.batch_norm(name='d_bn3') self.d_bn4 = ops.batch_norm(name='d_bn4')
def __init__(self, options): self.options = options # Used for sound embedding : No need Now!! self.g_s_bn0 = ops.batch_norm(name='g_s_bn0') self.g_s_bn1 = ops.batch_norm(name='g_s_bn1') self.g_s_bn2 = ops.batch_norm(name='g_s_bn2') self.d_s_bn0 = ops.batch_norm(name='d_s_bn0') self.d_s_bn1 = ops.batch_norm(name='d_s_bn1') self.d_s_bn2 = ops.batch_norm(name='d_s_bn2') # Used for noise vector self.g_bn0 = ops.batch_norm(name='g_bn0') self.g_bn1 = ops.batch_norm(name='g_bn1') self.g_bn2 = ops.batch_norm(name='g_bn2') self.g_bn3 = ops.batch_norm(name='g_bn3') self.d_bn1 = ops.batch_norm(name='d_bn1') self.d_bn2 = ops.batch_norm(name='d_bn2') self.d_bn3 = ops.batch_norm(name='d_bn3') self.d_bn4 = ops.batch_norm(name='d_bn4') self.d_bn5 = ops.batch_norm(name='d_bn5') self.down_bn1 = ops.batch_norm(name='down_bn1') #The batch normalization layers of the residual blocks self.r_bn0 = ops.batch_norm(name='res_bn0') self.r_bn1 = ops.batch_norm(name='res_bn1') self.r_bn3 = ops.batch_norm(name='res_bn3') self.r_bn4 = ops.batch_norm(name='res_bn4') self.r_bn6 = ops.batch_norm(name='res_bn6') self.r_bn7 = ops.batch_norm(name='res_bn7') self.r_bn9 = ops.batch_norm(name='res_bn9') self.r_bn10 = ops.batch_norm(name='res_bn10') self.g2_bn0 = ops.batch_norm(name='g2_bn0') self.g2_bn1 = ops.batch_norm(name='g2_bn1') self.g2_bn2 = ops.batch_norm(name='g2_bn2') self.d2_bn1 = ops.batch_norm(name='d2_bn1') self.d2_bn2 = ops.batch_norm(name='d2_bn2') self.d2_bn3 = ops.batch_norm(name='d2_bn3') self.d2_bn4 = ops.batch_norm(name='d2_bn4')