Unet-train_UNet_Unet去噪_u-net图像分割_u-net_Unet-train

# -*- coding: utf-8 -*- """ Created on Fri Apr 24 13:29:43 2020 @author: 46953 """ import os import tensorflow as tf from datetime import datetime from PIL import Image import matplotlib.pyplot as plt import numpy as np import skimage.io as io from skimage.transform import resize from skimage import transform,data from PIL import Image import numpy as np def read_data(data_dir): datas = [] labels = [] fpaths = [] for fname in os.listdir(data_dir): fpath = os.path.join(data_dir, fname) fpaths.append(fpath) image = Image.open(fpath) data = np.array(image) / 255.0 label = int(fname.split("_")[0]) data = resize(data, [128, 128,1], mode = 'constant') datas.append(data) labels.append(label) datas = np.array(datas) labels = np.array(labels) print("shape of datas: {}\tshape of labels: {}".format(datas.shape, labels.shape)) return datas x_train = tf.placeholder(dtype=tf.float32, shape=[None, 128, 128, 1], name="x_train") y_train = tf.placeholder(dtype=tf.float32, shape=[None, 128, 128, 1], name="y_train") lr = tf.placeholder(dtype=tf.float32) conv1 = tf.layers.conv2d(inputs=x_train, filters=16, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv1 = tf.layers.dropout(inputs=conv1, rate=0.1) conv1 = tf.layers.conv2d(inputs=conv1, filters=16, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) p1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=2, strides=2, padding="valid") conv2 = tf.layers.conv2d(inputs=p1, filters=32, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv2 = tf.layers.dropout(inputs=conv2, rate=0.1) conv2 = tf.layers.conv2d(inputs=conv2, filters=32, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) p2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=2, strides=2, padding="valid") conv3 = tf.layers.conv2d(inputs=p2, filters=64, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv3 = tf.layers.dropout(inputs=conv3, rate=0.1) conv3 = tf.layers.conv2d(inputs=conv3, filters=64, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) p3 = tf.layers.max_pooling2d(inputs=conv3, pool_size=2, strides=2, padding="valid") conv4 = tf.layers.conv2d(inputs=p3, filters=128, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv4 = tf.layers.dropout(inputs=conv4, rate=0.1) conv4 = tf.layers.conv2d(inputs=conv4, filters=128, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) p4 = tf.layers.max_pooling2d(inputs=conv4, pool_size=2, strides=2, padding="valid") conv5 = tf.layers.conv2d(inputs=p4, filters=256, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv5 = tf.layers.dropout(inputs=conv5, rate=0.2) conv5 = tf.layers.conv2d(inputs=conv5, filters=256, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) up1 = tf.layers.conv2d_transpose(inputs=conv5, filters=128, kernel_size=2, strides=2, padding="same", kernel_initializer=tf.initializers.glorot_uniform()) up1 = tf.concat([up1, conv4], axis=3) conv6 = tf.layers.conv2d(inputs=up1, filters=128, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv6 = tf.layers.dropout(inputs=conv6, rate=0.2) conv6 = tf.layers.conv2d(inputs=conv6, filters=128, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) up2 = tf.layers.conv2d_transpose(inputs=conv6, filters=64, kernel_size=2, strides=2, padding="same", kernel_initializer=tf.initializers.glorot_uniform()) up2 = tf.concat([up2, conv3], axis=3) conv7 = tf.layers.conv2d(inputs=up2, filters=64, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv7 = tf.layers.dropout(inputs=conv7, rate=0.2) conv7 = tf.layers.conv2d(inputs=conv7, filters=64, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) up3 = tf.layers.conv2d_transpose(inputs=conv7, filters=32, kernel_size=2, strides=2, padding="same", kernel_initializer=tf.initializers.glorot_uniform()) up3 = tf.concat([up3, conv2], axis=3) conv8 = tf.layers.conv2d(inputs=up3, filters=32, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv8 = tf.layers.dropout(inputs=conv8, rate=0.2) conv8 = tf.layers.conv2d(inputs=conv8, filters=32, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) up4 = tf.layers.conv2d_transpose(inputs=conv8, filters=16, kernel_size=2, strides=2, padding="same", kernel_initializer=tf.initializers.glorot_uniform()) up4 = tf.concat([up4, conv1], axis=3) conv9 = tf.layers.conv2d(inputs=up4, filters=16, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv9 = tf.layers.dropout(inputs=conv9, rate=0.1) conv9 = tf.layers.conv2d(inputs=conv9, filters=16, kernel_size=3, strides=1, padding="same", activation=tf.nn.relu, kernel_initializer=tf.initializers.glorot_uniform()) conv10 = tf.layers.conv2d(inputs=conv9, filters=1, kernel_size=1, strides=1, kernel_initializer=tf.initializers.glorot_uniform()) logits = tf.nn.sigmoid(conv10) learning_rate = 0.0051 y_out = logits cost = tf.reduce_sum(tf.square(y_out - y_train)) update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS) with tf.control_dependencies(update_ops): train_step = tf.train.AdamOptimizer(learning_rate).minimize(cost) # Create a saver for writing training checkpoints. evaluate=tf.reduce_sum(tf.square(y_out - y_train)) saver = tf.train.Saver(max_to_keep=30) config = tf.ConfigProto() config.gpu_options.allow_growth=True sess = tf.Session(config=config) sess.run(tf.initialize_all_variables()) noise_path = "D:/train" truth_path = "D:/truth" test_t_path = "D:/test_truth" test_n_path = "D:/test" Train= read_data(noise_path) Truth= read_data(truth_path) Train_vali=read_data(test_n_path) Truth_vali=read_data(test_t_path) channels=1 layers=4 filter_size=3 pool_size=2 features_root=16 keep_prob=1.0 n_class=1 s1=128 s2=128