卷积自编码去噪-tensorflow实现

from tensorflow.examples.tutorials.mnist import input_data import tensorflow as tf import numpy as np import utils mnist = input_data.read_data_sets("MNIST_data/", one_hot=True) sess = tf.InteractiveSession() batch_size = 256 noise_scale = 0.5 def weights_variable(shape): initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial) def bias_variable(shape): initial = tf.constant(0.1, shape=shape) return tf.Variable(initial) #strides 第一个参数：batch上的步长，第二个：height上的步长，第三个：weights上的步长，第四个：channel上的步长 def conv2d(x, W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME') def deconv2d(x, W, output): return tf.nn.conv2d_transpose(x, W, output, strides=[1, 2, 2, 1], padding='SAME') #ksize 第一个参数：batch上的池化，第二个：height上的池化，第三个：weights上的池化，第四个：channel上的池化 def max_pool_2x2(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') x = tf.placeholder(tf.float32, [None, 784], name='input') x_noise = tf.placeholder(tf.float32, [None, 784], name='input') #y_ = tf.placeholder(tf.float32, [None, 10]) #-1代表样本数量不固定，最后的1代表颜色通道数量 x_image = tf.reshape(x, [-1, 28, 28, 1]) x_image_noise = tf.reshape(x_noise, [-1, 28, 28, 1]) #编码 W_conv1 = weights_variable([3, 3, 1, 64]) b_conv1 = bias_variable([64]) h_conv1 = tf.nn.relu(conv2d(x_image_noise, W_conv1) + b_conv1, name='noise_layer') h_pool1 = max_pool_2x2(h_conv1) W_conv2 = weights_variable([3, 3, 64, 64]) b_conv2 = bias_variable([64]) h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) h_pool2 = max_pool_2x2(h_conv2) W_conv3 = weights_variable([3, 3, 64, 32]) b_conv3 = bias_variable([32]) h_conv3 = tf.nn.relu(conv2d(h_pool2, W_conv3) + b_conv3) h_pool3 = max_pool_2x2(h_conv3) #解码 W_deconv1 = weights_variable([3, 3, 32, 32]) h_deconv1 = deconv2d(h_pool3, W_deconv1, [batch_size, 7, 7, 32]) W_deconv2 = weights_variable([3, 3, 64, 32]) h_deconv2 = deconv2d(h_deconv1, W_deconv2, [batch_size, 14, 14, 64]) W_deconv3 = weights_variable([3, 3, 64, 64]) h_deconv3 = deconv2d(h_deconv2, W_deconv3, [batch_size, 28, 28, 64]) #卷积层 W_conv_final = weights_variable([3, 3, 64, 1]) b_conv_final = bias_variable([1]) h_conv_final = tf.nn.bias_add(conv2d(h_deconv3, W_conv_final), b_conv_final, name='output_layer') #输出图像 output_img = tf.reshape(h_conv_final, shape=[-1, 28, 28, 1]) output_img_fomat = utils.convert2int(output_img) output = tf.reshape(h_conv_final, shape=[-1, 784]) input = tf.reshape(x, shape=[-1, 784]) #训练 cost = tf.reduce_mean(tf.pow(tf.subtract(output, input), 2.0)) #计算平方误差 train_steps = tf.train.AdamOptimizer(0.001).minimize(cost) with tf.name_scope('images'): tf.summary.image('input', x_image, 1) tf.summary.image('gaussian', x_image_noise, 1) tf.summary.image('reconstruction', (output_img_fomat), 1) merged = tf.summary.merge_all() n_samples = int(mnist.train.num_examples) print('train samples: %d' % n_samples) print('batch size: %d' % batch_size) total_batch = int(n_samples / batch_size) print('total batchs: %d' % total_batch) init = tf.global_variables_initializer() sess.run(init) train_epochs = 35 log_dir = 'logs/mnist_with_summaries' test_writer = tf.summary.FileWriter(log_dir + '/test') saver = tf.train.Saver() for epoch in range(train_epochs): for batch_index in range(total_batch): batch_xs, batch_ys = mnist.train.next_batch(batch_size) noise_x = batch_xs + noise_scale * np.random.randn(*batch_xs.shape) noise_x = np.clip(noise_x, 0., 1.) _, train_loss, summaries = sess.run([train_steps, cost, merged], feed_dict={x: batch_xs, x_noise: noise_x}) print("epoch: %04d\tbatch: %04d\ttrain loss: %.9f" % (epoch + 1, batch_index + 1, train_loss)) test_writer.add_summary(summaries, epoch * total_batch + batch_index) saver.save(sess, './checkpoint_dir/CAEmodel') test_writer.close() n_test_samples = int(mnist.test.num_examples) test_total_batch = int(n_test_samples / batch_size)