Tensorflow2.0 卷积操作

卷积操作

"""
    keras.Conv2D 卷积操作与
    tf.nn.Conv2D 卷积操作
"""
import tensorflow as tf
from tensorflow.keras import layers

# 获取物理GPU的个数
gpus = tf.config.experimental.list_physical_devices("GPU")
for gpu in gpus:
    # 设置内存增长方式 自增长
    tf.config.experimental.set_memory_growth(gpu, True)
print("物理GPU个数:", len(gpus))

x = tf.random.normal([1, 32, 32, 3])

# 卷积核的个数 = 4 (决定了 输出维度) 卷积核大小 5*5 步长=1 不做padding
layer = layers.Conv2D(4, kernel_size=5, strides=1, padding="valid")
out = layer(x)
# [1, 32, 32, 3] ==> [1, 28, 28, 4] (4 由卷积核个数决定)
print(out.shape)

# 卷积核的个数 = 4 卷积核大小 5*5 步长=1 使得维度相同的 padding
layer = layers.Conv2D(4, kernel_size=5, strides=1, padding="same")
out = layer(x)
# [1, 32, 32, 3] ==> [1, 32, 32, 4]
print(out.shape)

# 卷积核的个数 = 4 卷积核大小 5*5 步长=2 使得维度相同的 padding
layer = layers.Conv2D(4, kernel_size=5, strides=2, padding="same")
out = layer(x)
# [1, 32, 32, 3] ==> [1, 16, 16, 4]
print(out.shape)

# 等同于 layer(x)  重写了 __call__魔法函数
print(layer.call(x).shape)

""" 
    权重与偏置  weight bias
"""
# 卷积核权重
# [5, 5, 3, 4]   [size.x, size.y, channel(通道), number(卷积个数)]
print(layer.kernel.shape)

# 偏置
# [4,]
print(layer.bias)

"""
    tf.nn.conv2d  更加底层 需要自己管理 权重 w + 偏置b 
"""
# 需要根据输入设置 权重维度
w = tf.random.normal([5, 5, 3, 4])
b = tf.zeros([4])
print(x.shape)

# 卷积操作  x:输入 w:权重  strides:步长  padding:周边是否填充
out = tf.nn.conv2d(x, w, strides=1, padding="VALID")
print(out.shape)

# 偏置的添加
out += b
print(out.shape)

# 卷积操作  x:输入 w:权重  strides:步长  padding:周边是否填充
out = tf.nn.conv2d(x, w, strides=2, padding="VALID")
print(out.shape)