def show_feature_map():
cat = plt.imread('cat.jpg') #unit8
plt.imshow(cat)
cat = tf.cast(cat, tf.float32) #[360, 300, 3]
x = tf.reshape(cat, [1, 360, 300, 3]) #[1, 360, 300, 3]
out = 25
with tf.variable_scope('conv1'):
w = tools.weight([3, 3, 3, out], is_uniform=True)
x_w = tf.nn.conv2d(x, w, strides=[1, 1, 1, 1], padding='SAME')
b = tools.bias([out])
x_b = tf.nn.bias_add(x_w, b)
x_relu = tf.nn.relu(x_b)
n_feature = int(x_w.get_shape()[-1])
sess = tf.Session()
sess.run(tf.global_variables_initializer())
feature_map = tf.reshape(x_w, [360, 300, out])
images = tf.image.convert_image_dtype(feature_map, dtype=tf.uint8)
images = sess.run(images)
plt.figure(figsize=(10, 10))
for i in np.arange(0, n_feature):
plt.subplot(5, 5, i + 1)
plt.axis('off')
plt.imshow(images[:, :, i])
plt.show()
import tensorflow as tf
import matplotlib.pyplot as plt
import tools
cat = plt.imread('cat.jpg') #unit8
plt.imshow(cat)
cat = tf.cast(cat, tf.float32) #[360, 300, 3]
x = tf.reshape(cat, [1, 360, 300, 3]) #[1, 360, 300, 3]
# First conv
with tf.variable_scope('conv1'):
w = tools.weight([3, 3, 3, 16], is_uniform=True)
x_w = tf.nn.conv2d(x, w, strides=[1, 1, 1, 1], padding='SAME')
b = tools.bias([16])
x_b = tf.nn.bias_add(x_w, b)
x_relu = tf.nn.relu(x_b)
x_pool = tools.pool('test1',
x_relu,
kernel=[1, 2, 2, 1],
stride=[1, 2, 2, 1],
is_max_pool=True)
# Second conv
with tf.variable_scope('conv2'):
w2 = tools.weight([3, 3, 16, 32], is_uniform=True)
x_w2 = tf.nn.conv2d(x_pool, w2, strides=[1, 1, 1, 1], padding='SAME')
b2 = tools.bias([32])
第一层：
"""
# 卷积层conv1
with tf.name_scope('conv1'):
with tf.name_scope('kernel1'):
kernel1 = tools.Weight_with_WeightLoss(shape=[5, 5, 3, 64],
stddev=5e-2,
lamda=0)
tools.variables_summaries(kernel1, 'conv1/kernel1')
with tf.name_scope('conv_op'):
conv1_result = tf.nn.conv2d(image_holder,
kernel1,
strides=[1, 1, 1, 1],
padding='SAME')
with tf.name_scope('bias_op'):
b1 = tools.bias(0.1, shape=[64])
tools.variables_summaries(b1, 'conv1/b1')
with tf.name_scope('activate'):
act1_result = tf.nn.relu(tf.nn.bias_add(conv1_result, b1))
# 池化层与局部响应归一化
with tf.name_scope('max_pool_and_norm1'):
with tf.name_scope('max_pool'):
pool1_result = tf.nn.max_pool(act1_result,
ksize=[1, 2, 2, 1],
strides=[1, 3, 3, 1],
padding='SAME')
with tf.name_scope('lrn1'):
norm1_result = tf.nn.lrn(pool1_result,
4,
bias=1.0,
import tools
#%%
cat = plt.imread('cat.jpg') #unit8
plt.imshow(cat)
cat = tf.cast(cat, tf.float32) #[360, 300, 3]
x = tf.reshape(cat, [1, 360, 300, 3]) #[1, 360, 300, 3]
#%%
# First conv
with tf.variable_scope('conv1'):
w = tools.weight([3,3,3,16], is_uniform=True)
x_w = tf.nn.conv2d(x, w, strides=[1, 1, 1, 1], padding='SAME')
b = tools.bias([16])
x_b = tf.nn.bias_add(x_w, b)
x_relu = tf.nn.relu(x_b)
x_pool = tools.pool('test1', x_relu, kernel=[1,2,2,1], stride=[1,2,2,1],is_max_pool=True)
# Second conv
with tf.variable_scope('conv2'):
w2 = tools.weight([3,3,16,32], is_uniform=True)
x_w2 = tf.nn.conv2d(x_pool, w2, strides=[1, 1, 1, 1], padding='SAME')
b2 = tools.bias([32])
x_b2 = tf.nn.bias_add(x_w2, b2)
x_relu2 = tf.nn.relu(x_b2)