["conv", [3, 3, 128, 128], [1, 1, 1, 1], 0.1, "SAME", False],
["conv", [3, 3, 128, 128], [1, 1, 1, 1], 0.1, "SAME", False],
["conv", [3, 3, 128, 128], [1, 1, 1, 1], 0.1, "SAME", False],
["avgpool", [1, 8, 8, 1], [1, 4, 4, 1]],
#["dropout",keep_prob_holder]
]
line_shape_result, len_shape_length = net.CNN_Interface(
image,
CNN_LAYERS,
bias=False,
active_function=tf.nn.relu,
output_each_layer=True)
Layer[0] = len_shape_length
label = net.FC_Interface(line_shape_result,
Layer,
keep_prob_layer=keep_prob_holder,
stddev=0.001,
active_function=tf.nn.relu)
with tf.variable_scope("Sample_Output"):
#测试输出
label_to_num = tf.argmax(label, 1)[0]
label__to_num = tf.argmax(label_, 1)[0]
label_sampel = tf.nn.softmax(label[0])
with tf.variable_scope("Train_Model"):
#总计训练次数
global_step = tf.Variable(0, trainable=False)
#学习率
LEARNING_RATE_BASE = tf.Variable(LEARNING_RATE_BASE)
#学习率指数衰减
