unet如何向网络中导入数据 代码示范

以下是用Python代码示范如何向UNET网络中导入数据：

import os
import numpy as np
from keras.preprocessing.image import load_img, img_to_array

# 定义数据路径
data_path = "/path/to/data"

# 定义图像大小
img_rows = 256
img_cols = 256

# 定义训练集和验证集的比例
train_val_ratio = 0.8

# 定义训练集和验证集的图像和标签列表
train_images = []
train_labels = []
val_images = []
val_labels = []

# 遍历数据路径下的所有图像和标签
for filename in os.listdir(data_path):
    if filename.endswith(".jpg"):
        img_path = os.path.join(data_path, filename)
        label_path = os.path.join(data_path, filename[:-4] + "_mask.png")

        # 加载图像和标签
        img = load_img(img_path, target_size=(img_rows, img_cols))
        label = load_img(label_path, target_size=(img_rows, img_cols), grayscale=True)

        # 将图像和标签转换为数组
        img = img_to_array(img)
        label = img_to_array(label)

        # 归一化图像和标签
        img = img / 255.0
        label = label / 255.0

        # 将图像和标签添加到训练集或验证集列表中
        if np.random.rand() < train_val_ratio:
            train_images.append(img)
            train_labels.append(label)
        else:
            val_images.append(img)
            val_labels.append(label)

# 将训练集和验证集转换为数组
train_images = np.array(train_images)
train_labels = np.array(train_labels)
val_images = np.array(val_images)
val_labels = np.array(val_labels)

# 打印数据集的形状
print("Train images shape:", train_images.shape)
print("Train labels shape:", train_labels.shape)
print("Val images shape:", val_images.shape)
print("Val labels shape:", val_labels.shape)

以上代码将从指定路径下加载所有图像和标签，将它们转换为数组，并归一化它们。然后，它将随机将它们分为训练集和验证集，并将它们转换为数组。最后，它将打印数据集的形状。在实际应用中，您可能需要根据自己的数据集的特点对代码进行一些微调。