input-data-V1.py

# -*- coding:utf-8 -*-
import glob
import os
import numpy as np
import tensorflow as tf

from tensorflow.python.platform import gfile


# fornix_L ILF_L Cingulum_hippocampus_L ATR_L
INPUT_DATA = 'D:/python/deep-learning/MRI-2D/BMP-cnn/BMP-data/ATR_L'
# 以numpy的格式保存图片数据
OUTPUT_FILE = 'D:/python/deep-learning/MRI-2D/data_npy/data_ATR_L.npy'


def create_image_list(sess):
    sub_dirs = [x[0] for x in os.walk(INPUT_DATA)][1:]  # sub_dirs是所有子文件名

    training_images = []
    training_labels = []
    testing_images = []
    testing_labels = []
    current_label = 0
    chance = 0  # 用来分测试集和训练集

    for sub_dir in sub_dirs:

        extensions = ['bmp']
        file_list = []
        dir_name = os.path.basename(sub_dir)  # os.path.basename(),返回path最后的文件名
        for extension in extensions:
            file_glob = os.path.join(INPUT_DATA, dir_name, '*.' + extension)
            file_list.extend(glob.glob(file_glob))
        if not file_list: continue
        print(len(file_list))  # 把一个子文件里的所有图片加入到file_list里面。

        for file_name in file_list:
            image_raw_data = gfile.FastGFile(file_name, 'rb').read()
            image = tf.image.decode_bmp(image_raw_data, channels=3)
            image = tf.image.convert_image_dtype(image, dtype=tf.float32)
            image = tf.image.resize_images(image, [299, 299],  method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
            # raw_data=[46,810]=37260
            # 问题出在inception-V3 只认299*299*3的彩色图片
            image_value = sess.run(image)
            chance = chance + 1  # chance是处理图片的数量

            if chance < 273:
                training_images.append(image_value)
                training_labels.append(current_label)
            elif 272 < chance < 337:
                testing_images.append(image_value)
                testing_labels.append(current_label)
            elif 336 < chance < 689:
                training_images.append(image_value)
                training_labels.append(current_label)
            elif 688 < chance < 769:
                testing_images.append(image_value)
                testing_labels.append(current_label)

        current_label += 1

    # 随机打乱训练数据
    state = np.random.get_state()
    np.random.shuffle(training_images)
    np.random.set_state(state)
    np.random.shuffle(training_labels)

    state = np.random.get_state()
    np.random.shuffle(testing_images)
    np.random.set_state(state)
    np.random.shuffle(testing_labels)

    return np.asarray([training_images, training_labels, testing_images, testing_labels])


def main():
    with tf.Session() as sess:
        processed_data = create_image_list(sess)
        np.save(OUTPUT_FILE, processed_data)


if __name__ == '__main__':
    main()