tensorflow53 《面向机器智能的TensorFlow实战》笔记-05-01 卷积基础

01 get_shape

# 《面向机器智能的Tensor Flow实战》05 目标识别与分类
# win10 Tensorflow-gpu1.1.0 python3.5.3
# CUDA v8.0 cudnn-8.0-windows10-x64-v5.1
# 原书代码(tensorflow0.8)：https://github.com/backstopmedia/tensorflowbook
# tensorflow不同版本api变化:https://github.com/tensorflow/tensorflow/releases
# filename:tfmi05.01.py # get_shape

import tensorflow as tf
import numpy as np
sess = tf.InteractiveSession()
image_batch = tf.constant([
        [  # First Image
            [[0, 255, 0], [0, 255, 0], [0, 255, 0]],
            [[0, 255, 0], [0, 255, 0], [0, 255, 0]]
        ],
        [  # Second Image
            [[0, 0, 255], [0, 0, 255], [0, 0, 255]],
            [[0, 0, 255], [0, 0, 255], [0, 0, 255]]
        ]
    ])
image_batch.get_shape()
sess.run(image_batch)[0][0][0]
print(image_batch) # Tensor("Const:0", shape=(2, 2, 3, 3), dtype=int32)
print(sess.run(image_batch)[0][0][0]) # [  0 255   0]

02 卷积

# 《面向机器智能的Tensor Flow实战》05 目标识别与分类
# win10 Tensorflow-gpu1.1.0 python3.5.3
# CUDA v8.0 cudnn-8.0-windows10-x64-v5.1
# 原书代码(tensorflow0.8)：https://github.com/backstopmedia/tensorflowbook
# tensorflow不同版本api变化:https://github.com/tensorflow/tensorflow/releases
# filename:tfmi05.02.py # 卷积

import tensorflow as tf
import numpy as np
sess = tf.InteractiveSession()
input_batch = tf.constant([
        [  # First Input
            [[0.0], [1.0]],
            [[2.0], [3.0]]
        ],
        [  # Second Input
            [[2.0], [4.0]],
            [[6.0], [8.0]]
        ]
    ])
kernel = tf.constant([
        [
            [[1.0, 2.0]]
        ]
    ])
conv2d = tf.nn.conv2d(input_batch, kernel, strides=[1, 1, 1, 1], padding='SAME')
sess.run(conv2d)
print(conv2d.eval())
'''
[[[[  0.   0.]
   [  1.   2.]]
  [[  2.   4.]
   [  3.   6.]]]
 [[[  2.   4.]
   [  4.   8.]]
  [[  6.  12.]
   [  8.  16.]]]]
'''

lower_right_image_pixel = sess.run(input_batch)[0][1][1]
lower_right_kernel_pixel = sess.run(conv2d)[0][1][1]
lower_right_image_pixel, lower_right_kernel_pixel
print("lower_right_image_pixel: ", lower_right_image_pixel)
print("lower_right_kernel_pixel: ", lower_right_kernel_pixel)
print("lower_right_image_pixel: ", lower_right_image_pixel)
'''
lower_right_image_pixel:  [ 3.]
lower_right_kernel_pixel:  [ 3.  6.]
lower_right_image_pixel:  [ 3.]
'''

input_batch = tf.constant([
        [  # First Input (6x6x1)
            [[0.0], [1.0], [2.0], [3.0], [4.0], [5.0]],
            [[0.1], [1.1], [2.1], [3.1], [4.1], [5.1]],
            [[0.2], [1.2], [2.2], [3.2], [4.2], [5.2]],
            [[0.3], [1.3], [2.3], [3.3], [4.3], [5.3]],
            [[0.4], [1.4], [2.4], [3.4], [4.4], [5.4]],
            [[0.5], [1.5], [2.5], [3.5], [4.5], [5.5]],
        ],
    ])

kernel = tf.constant([  # Kernel (3x3x1)
        [[[0.0]], [[0.5]], [[0.0]]],
        [[[0.0]], [[1.0]], [[0.0]]],
        [[[0.0]], [[0.5]], [[0.0]]]
    ])

# NOTE: the change in the size of the strides parameter.
conv2d = tf.nn.conv2d(input_batch, kernel, strides=[1, 3, 3, 1], padding='SAME')
sess.run(conv2d)
print(conv2d.eval())
'''
[[[[ 2.20000005]
   [ 8.19999981]]
  [[ 2.79999995]
   [ 8.80000019]]]]
'''

import matplotlib as mil
#mil.use('svg')
mil.use("nbagg")
from matplotlib import pyplot
fig = pyplot.gcf()
fig.set_size_inches(4, 4)

03 激活函数/池化/归一化/高级层

# 《面向机器智能的Tensor Flow实战》05 目标识别与分类
# win10 Tensorflow-gpu1.1.0 python3.5.3
# CUDA v8.0 cudnn-8.0-windows10-x64-v5.1
# 原书代码(tensorflow0.8)：https://github.com/backstopmedia/tensorflowbook
# tensorflow不同版本api变化:https://github.com/tensorflow/tensorflow/releases
# filename:tfmi05.03.py # 激活函数/池化/归一化/高级层

import tensorflow as tf
import numpy as np
sess = tf.InteractiveSession()
features = tf.range(-2, 3)
# Keep note of the value for negative features
sess.run([features, tf.nn.relu(features)])
print([features, tf.nn.relu(features)]) # [<tf.Tensor 'range:0' shape=(5,) dtype=int32>, <tf.Tensor 'Relu_1:0' shape=(5,) dtype=int32>]
print(features.eval(), tf.nn.relu(features).eval()) # [-2 -1  0  1  2] [0 0 0 1 2]

features = tf.to_float(tf.range(-1, 3))
sess.run([features, tf.sigmoid(features)])
print(features.eval(), tf.sigmoid(features).eval()) # [-1.  0.  1.  2.] [ 0.26894143  0.5         0.7310586   0.88079703]

features = tf.to_float(tf.range(-1, 3))
sess.run([features, tf.tanh(features)])
print(features.eval(), tf.tanh(features).eval()) # [-1.  0.  1.  2.] [-0.76159418  0.          0.76159418  0.96402758]

features = tf.constant([-0.1, 0.0, 0.1, 0.2])
sess.run([features, tf.nn.dropout(features, keep_prob=0.5)])
print(features.eval(), tf.nn.dropout(features, keep_prob=0.5).eval()) # [-0.1  0.   0.1  0.2] [-0. 0. 0. 0.40000001]

batch_size=1
input_height = 3
input_width = 3
input_channels = 1
layer_input = tf.constant([
        [
            [[1.0], [0.2], [1.5]],
            [[0.1], [1.2], [1.4]],
            [[1.1], [0.4], [0.4]]
        ]
    ])
# The strides will look at the entire input by using the image_height and image_width
kernel = [batch_size, input_height, input_width, input_channels]
max_pool = tf.nn.max_pool(layer_input, kernel, [1, 1, 1, 1], "VALID")
sess.run(max_pool)
print(max_pool.eval()) # [[[[ 1.5]]]]

batch_size=1
input_height = 3
input_width = 3
input_channels = 1
layer_input = tf.constant([
        [
            [[1.0], [1.0], [1.0]],
            [[1.0], [0.5], [0.0]],
            [[0.0], [0.0], [0.0]]
        ]
    ])
# The strides will look at the entire input by using the image_height and image_width
kernel = [batch_size, input_height, input_width, input_channels]
max_pool = tf.nn.avg_pool(layer_input, kernel, [1, 1, 1, 1], "VALID")
sess.run(max_pool)
print(max_pool) # Tensor("AvgPool:0", shape=(1, 1, 1, 1), dtype=float32)
print(max_pool.eval()) # [[[[ 0.5]]]]


layer_input = tf.constant([
        [[[ 1.]], [[ 2.]], [[ 3.]]]
    ])
lrn = tf.nn.local_response_normalization(layer_input)
sess.run([layer_input, lrn])
print(layer_input.eval(), lrn.eval())
'''
[[[[ 1.]]
  [[ 2.]]
  [[ 3.]]]] [[[[ 0.70710677]]
  [[ 0.89442718]]
  [[ 0.94868326]]]]
'''

image_input = tf.constant([
            [
                [[0., 0., 0.], [255., 255., 255.], [254., 0., 0.]],
                [[0., 191., 0.], [3., 108., 233.], [0., 191., 0.]],
                [[254., 0., 0.], [255., 255., 255.], [0., 0., 0.]]
            ]
        ])
conv2d = tf.contrib.layers.convolution2d(
    image_input,
    num_outputs=4,
    kernel_size=(1,1),          # It's only the filter height and width.
    activation_fn=tf.nn.relu,
    stride=(1, 1),              # Skips the stride values for image_batch and input_channels.
    trainable=True)

# It's required to initialize the variables used in convolution2d's setup.
sess.run(tf.global_variables_initializer())
sess.run(conv2d)
print(conv2d.eval())
'''
[[[[   0.            0.            0.            0.        ]
   [   9.16278458    0.           56.88573456    0.        ]
   [   0.           87.55300903    0.           11.87749958]]

  [[ 130.84643555    0.           48.07785034    0.        ]
   [  75.74397278    0.          106.45916748    0.        ]
   [ 130.84643555    0.           48.07785034    0.        ]]

  [[   0.           87.55300903    0.           11.87749958]
   [   9.16278458    0.           56.88573456    0.        ]
   [   0.            0.            0.            0.        ]]]]
'''
features = tf.constant([[[1.2], [3.4]]])
fc = tf.contrib.layers.fully_connected(features, num_outputs=2)
sess.run(tf.global_variables_initializer())
sess.run(fc)
print(fc.eval())
'''
[[[ 0.24772798  0.        ]
  [ 0.70189595  0.        ]]]
'''