def __init__(self, in_shape):
super(Discriminator, self).__init__()
"""
反卷积和dense层采用偏置 各自2参数
2+2+2=6 一共六个参数个数(指独立大参数self.w self.b的个数)
"""
self.Conv2d_1 = my_layers.Conv2D(input_shape=in_shape,
out_depth=64,
filter_size=[5, 5],
strides=[2, 2],
use_bias=True,
pandding_way="SAME")
self.LeakyReLU_1 = my_layers.LeakyReLU(
in_shape=self.Conv2d_1.out_shape)
self.DropOut_1 = my_layers.Dropout(in_shape=self.LeakyReLU_1.out_shape,
dropout_rate=0.3)
self.Conv2d_2 = my_layers.Conv2D(input_shape=self.DropOut_1.out_shape,
out_depth=128,
filter_size=[5, 5],
strides=[2, 2],
use_bias=True,
pandding_way="SAME")
self.LeakyReLU_2 = my_layers.LeakyReLU(
in_shape=self.Conv2d_2.out_shape)
self.DropOut_2 = my_layers.Dropout(in_shape=self.LeakyReLU_2.out_shape,
dropout_rate=0.3)
next_shape = 1
for i in self.DropOut_2.out_shape:
next_shape *= i
self.Dense = my_layers.Dense(next_shape, units=1)
def __init__(self, in_dim):
super(Generator, self).__init__()
"""
bn层两个参数
反卷积和dense层不采用偏置 各自只有一个参数
1+2+1+2+1+2+1=10 一共十个参数个数(指独立大参数self.w self.b的个数)
"""
self.Dense_1 = my_layers.Dense(in_dim, 7 * 7 * 256, use_bias=False)
self.BacthNormalization_1 = my_layers.BatchNormalization(
in_shape=self.Dense_1.out_dim)
self.LeakyReLU_1 = my_layers.LeakyReLU(
in_shape=self.BacthNormalization_1.out_shape)
self.Conv2dTranspose_2 = my_layers.Conv2DTranspose(
in_shape=[7, 7, 256],
out_depth=128,
kernel_size=[5, 5],
strides=[1, 1],
pandding_way="SAME",
use_bias=False)
assert self.Conv2dTranspose_2.out_shape == [7, 7, 128]
self.BacthNormalization_2 = my_layers.BatchNormalization(
in_shape=self.Conv2dTranspose_2.out_shape)
self.LeakyReLU_2 = my_layers.LeakyReLU(
in_shape=self.BacthNormalization_2.out_shape)
self.Conv2dTranspose_3 = my_layers.Conv2DTranspose(
in_shape=self.LeakyReLU_2.out_shape,
out_depth=64,
kernel_size=[5, 5],
strides=[2, 2],
pandding_way="SAME",
use_bias=False)
assert self.Conv2dTranspose_3.out_shape == [14, 14, 64]
self.BacthNormalization_3 = my_layers.BatchNormalization(
in_shape=self.Conv2dTranspose_3.out_shape)
self.LeakyReLU_3 = my_layers.LeakyReLU(
in_shape=self.BacthNormalization_3.out_shape)
self.Conv2dTranspose_4 = my_layers.Conv2DTranspose(
in_shape=self.LeakyReLU_3.out_shape,
out_depth=1,
kernel_size=[5, 5],
strides=[2, 2],
pandding_way="SAME",
use_bias=False)
assert self.Conv2dTranspose_4.out_shape == [28, 28, 1]