Add exam.8

exam/step8/ResNetCompleted.py

+import tensorflow as tf
+
+def multiChannelWeightLayer(Inputs, batchNormTraining,batchSize):
+    '''
+    对输入完成BatchNorm + relu + Wx_plus_b操作
+    :param Inputs: 输入张量
+    :param batchNormTraining: batchNorm层Training参数，在训练和预测阶段传入不同值
+    :return:
+    '''
+
+    batchNorm = tf.layers.batch_normalization(Inputs, training=batchNormTraining)
+    relu = tf.nn.relu(batchNorm)
+    transposed = tf.transpose(relu, [0, 3, 1, 2])
+    num_channels = Inputs.get_shape()[-1].value
+    size = Inputs.get_shape()[1].value
+
+    weight = tf.Variable(tf.truncated_normal(shape=(size, size)), dtype=tf.float32, trainable=True)
+    weight_expand = tf.expand_dims(weight, axis=0)
+    weight_nchannels = tf.tile(weight_expand, tf.constant([num_channels, 1, 1]))
+    batch_expand = tf.expand_dims(weight_nchannels, axis=0)
+    weight_final = tf.tile(batch_expand, tf.concat([tf.stack([batchSize,1],axis=0),[1,1]],axis=0))
+
+    WX = tf.matmul(transposed, weight_final)
+
+    bias = tf.Variable(tf.truncated_normal(shape=(size,)), dtype=tf.float32, trainable=True)
+    bias_expand = tf.expand_dims(bias, axis=0)
+    bias_size = tf.tile(bias_expand, tf.constant([size, 1]))
+    bias_channels_expand = tf.expand_dims(bias_size, axis=0)
+    bias_channels = tf.tile(bias_channels_expand, tf.constant([num_channels, 1, 1]))
+    bias_batch_expand = tf.expand_dims(bias_channels, axis=0)
+    bias_final = tf.tile(bias_batch_expand,tf.concat([tf.stack([batchSize,1],axis=0),[1,1]],axis=0))
+
+    WX_PLUS_B = WX + bias_final
+
+    outputs = tf.transpose(WX_PLUS_B, [0, 2, 3, 1])
+
+    return outputs
+
+
+def ResNetBlock(Inputs, batchNormTraining, batchSize):
+    '''
+    堆叠两次（BatchNorm + relu + Wx_plus_b）操作，形成一个残差模块
+    :param Inputs: 输入张量
+    :param batchNormTraining: batchNorm层Training参数，在训练和预测阶段传入不同值
+    :return:
+    '''
+    shortcut = Inputs
+    wx_1 = multiChannelWeightLayer(Inputs, batchNormTraining=batchNormTraining, batchSize=batchSize)
+    res = multiChannelWeightLayer(wx_1, batchNormTraining=batchNormTraining, batchSize=batchSize)
+    outputs = tf.add(shortcut, res)
+
+    return outputs
+
+
+
+
+
+BNTraining = tf.placeholder(tf.bool)
+keeProb = tf.placeholder(tf.float32, shape=())
+batchImgInput = tf.placeholder(tf.float32, shape=(None, 224, 224, 3))
+labels = tf.placeholder(tf.float32, shape=(None, 4))
+InputBatchSize = tf.placeholder(tf.int32)
+
+
+conv1 = tf.layers.conv2d(batchImgInput, filters=96, kernel_size=11, strides=4,padding='same',
+                 activation=tf.nn.relu)
+pool1 = tf.layers.max_pooling2d(conv1, pool_size=2, strides=2, padding='same')
+conv2 = tf.layers.conv2d(pool1, filters=128, kernel_size=3, strides=1,padding='same',
+                 activation=tf.nn.relu)
+pool2 = tf.layers.max_pooling2d(conv2, pool_size=3, strides=2, padding='same')
+
+resBlock1 = ResNetBlock(pool2, batchNormTraining=BNTraining, batchSize=InputBatchSize)
+conv3 = tf.layers.conv2d(resBlock1, filters=128, kernel_size=3,strides=1,padding='same',
+                 activation=tf.nn.relu)
+resBlock2 = ResNetBlock(conv3, batchNormTraining=BNTraining, batchSize=InputBatchSize)
+conv4 =  tf.layers.conv2d(resBlock2, filters=64, kernel_size=3,strides=1,padding='same',
+                 activation=tf.nn.relu)
+resBlock3 = ResNetBlock(conv4, batchNormTraining=BNTraining, batchSize=InputBatchSize)
+conv5 = tf.layers.conv2d(resBlock3, filters=64, kernel_size=3,strides=1,padding='same',
+                 activation=tf.nn.relu)
+
+pool3 = tf.layers.max_pooling2d(conv5, pool_size=2, strides=2, padding='same')
+flattened = tf.layers.flatten(pool3)
+
+dense1 = tf.layers.dense(flattened, units=256)
+dropout1 = tf.nn.dropout(dense1, keeProb)
+outputs = tf.layers.dense(dropout1, units=4)
+
+loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=outputs, labels=labels))
+update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
+with tf.control_dependencies(update_ops):
+    train = tf.train.AdamOptimizer().minimize(loss)
+
+with tf.Session() as sess:
+    sess.run(tf.global_variables_initializer())
+    tf.train.export_meta_graph(filename='step8/modelInfo/ResNet',
+                               graph=tf.get_default_graph())
+
+tf.reset_default_graph()
\ No newline at end of file

exam/step8/ResNetForUsers.py

+# 先定义如下函数
+
+# multiChannelWeightLayer
+# 使用所学的tf的API, transpose()/expand_dims()/tile()等
+
+# ResNetBlock
+# 堆叠两个multiChannelWeightLayer, 然后将输出加上输入
+
+#********** Begin **********#
+
+import tensorflow as tf
+
+def multiChannelWeightLayer(Inputs, batchNormTraining,batchSize):
+    '''
+    对输入完成BatchNorm + relu + Wx_plus_b操作
+    :param Inputs: 输入张量
+    :param batchNormTraining: batchNorm层Training参数，在训练和预测阶段传入不同值
+    :return:
+    '''
+
+
+
+def ResNetBlock(Inputs, batchNormTraining, batchSize):
+    '''
+    堆叠两次（BatchNorm + relu + Wx_plus_b）操作，形成一个残差模块
+    :param Inputs: 输入张量
+    :param batchNormTraining: batchNorm层Training参数，在训练和预测阶段传入不同值
+    :return:
+    '''
+
+
+# 首先定义五个placeholder
+# BNTraining / keeProb / batchImgInput / labels / InputBatchSize
+
+# 模型结构:
+# (tf.layers.conv2d + tf.layers.max_pooling2d)*2 + (ResNetBlock + tf.layers.conv2d)*3
+# tf.layers.max_pooling2d + tf.layers.flatten + tf.layers.dense(unit = 256) + tf.nn.dropout + tf.layers.dense
+
+# 损失: tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2())
+# 优化器: tf.train.AdamOptimizer()
+
+
+# 将模型保存至step8/userModelInfo文件夹，保存名称为ResNet
+with tf.Session() as sess:
+    sess.run(tf.global_variables_initializer())
+    tf.train.export_meta_graph(filename='step8/userModelInfo/ResNet',
+                               graph=tf.get_default_graph())
+
+#********** End **********#

exam/step8/ResNetTest.py

+import os
+os.environ[\"TF_CPP_MIN_LOG_LEVEL\"]='3'
+
+import warnings
+warnings.filterwarnings('ignore')
+
+import ResNetCompleted
+import ResNetForUsers
+
+rightModelPath = 'step8/modelInfo/ResNet'
+userModelPath = 'step8/userModelInfo/ResNet'
+
+# print(os.path.exists(rightModelPath))
+# print(os.path.exists(userModelPath))
+# print(os.path.getsize(rightModelPath))
+# print(os.path.getsize(userModelPath))
+
+
+try:
+
+    isRight = os.path.getsize(rightModelPath)==os.path.getsize(userModelPath)
+    # print(isRight)
+
+    if isRight:
+        print('恭喜你通过本关测试!你已掌握ResNet结构!',end='')
+    else:
+        print('未能通过本关测试!模块结构有误!')
+except:
+    print('未能通过本关测试!模型结构文件没能正确保存!')
\ No newline at end of file