TensorFlow2-维度变换

TensorFlow2-维度变换

Outline（大纲）

• shape, ndim

• reshape

• expand_dims/squeeze

• transpose

图片视图

• [b, 28, 28] # 保存b张图片，28行，28列（保存数据一般行优先），图片的数据没有被破坏
• [b, 28*28] # 保存b张图片，不考虑图片的行和列，只保存图片的数据，不关注图片数据的细节
• [b, 2, 14*28] # 保存b张图片，把图片分为上下两个部分，两个部分具体多少行是不清楚的
• [b, 28, 28, 1] # 保存b张图片，28行，28列，1个通道

First Reshape（重塑视图）

import tensorflow as tf

a = tf.random.normal([4, 28, 28, 3])

a.shape, a.ndim

(TensorShape([4, 28, 28, 3]), 4)

tf.reshape(a, [4, 784, 3]).shape  # 给出一张图片某个通道的数据，丢失行、宽的信息

TensorShape([4, 784, 3])

tf.reshape(a, [4, -1, 3]).shape  # 4*(-1)*3 = 4*28*28*3

TensorShape([4, 784, 3])

tf.reshape(a, [4, 784*3]).shape  # 给出一张图片的所有数据，丢失行、宽和通道的信息

TensorShape([4, 2352])

tf.reshape(a, [4, -1]).shape

TensorShape([4, 2352])

Second Reshape（恢复视图）

tf.reshape(tf.reshape(a, [4, -1]), [4, 28, 28, 3]).shape

TensorShape([4, 28, 28, 3])

tf.reshape(tf.reshape(a, [4, -1]), [4, 14, 56, 3]).shape

TensorShape([4, 14, 56, 3])

tf.reshape(tf.reshape(a, [4, -1]), [4, 1, 784, 3]).shape

TensorShape([4, 1, 784, 3])

first reshape:

• images: [4,28,28,3]
• reshape to: [4,784,3]

second reshape:

• [4,784,3]  height:28,width:28  [4,28,28,3] √
• [4,784,3]  height:14,width:56  [4,14,56,3] ×
• [4,784,3]  width:28,height:28  [4,28,28,3] ×

Transpose（转置）

a = tf.random.normal((4, 3, 2, 1))

a.shape

TensorShape([4, 3, 2, 1])

tf.transpose(a).shape

TensorShape([1, 2, 3, 4])

tf.transpose(a, perm=[0, 1, 3, 2]).shape  # 按照索引替换维度

TensorShape([4, 3, 1, 2])

a = tf.random.normal([4, 28, 28, 3])  # b,h,w,c

a.shape

TensorShape([4, 28, 28, 3])

tf.transpose(a, [0, 2, 1, 3]).shape  # b,2,h,c

TensorShape([4, 28, 28, 3])

tf.transpose(a, [0, 3, 2, 1]).shape  # b,c,w,h

TensorShape([4, 3, 28, 28])

tf.transpose(a, [0, 3, 1, 2]).shape  # b,c,h,w

TensorShape([4, 3, 28, 28])

Expand_dims（增加维度）

• a:[classes, students, classes]

add school dim（增加学校的维度）:

• [1, 4, 35, 8] + [1, 4, 35, 8] = [2, 4, 35, 8]

a = tf.random.normal([4, 25, 8])

a.shape

TensorShape([4, 25, 8])

tf.expand_dims(a, axis=0).shape  # 索引0前

TensorShape([1, 4, 25, 8])

tf.expand_dims(a, axis=3).shape  # 索引3前

TensorShape([4, 25, 8, 1])

tf.expand_dims(a,axis=-1).shape  # 索引-1后

TensorShape([4, 25, 8, 1])

tf.expand_dims(a,axis=-4).shape  # 索引-4后，即左边空白处

TensorShape([1, 4, 25, 8])

Squeeze（挤压维度）

Only squeeze for shape = 1 dim（只删除维度为1的维度）

• [4, 35, 8, 1] = [4, 35, 8]
• [1, 4, 35, 8] = [14, 35, 8]
• [1, 4, 35, 1] = [4, 35, 8]

tf.squeeze(tf.zeros([1,2,1,1,3])).shape

TensorShape([2, 3])

a = tf.zeros([1,2,1,3])

a.shape

TensorShape([1, 2, 1, 3])

tf.squeeze(a,axis=0).shape

TensorShape([2, 1, 3])

tf.squeeze(a,axis=2).shape

TensorShape([1, 2, 3])

tf.squeeze(a,axis=-2).shape

TensorShape([1, 2, 3])

tf.squeeze(a,axis=-4).shape

TensorShape([2, 1, 3])