Question

我有一组训练图像，其结构如下：

private void initialize() {


        frame = new JFrame();
        frame.setBounds(100, 100, 577, 443);
        frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);

        try {
            UIManager.setLookAndFeel("com.seaglasslookandfeel.SeaGlassLookAndFeel");
        } catch (ClassNotFoundException | InstantiationException | IllegalAccessException
                | UnsupportedLookAndFeelException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }

        JTabbedPane tabbedPane = new JTabbedPane(JTabbedPane.TOP);
        frame.getContentPane().add(tabbedPane, BorderLayout.CENTER);



        JPanel panel = new JPanel();
        tabbedPane.addTab("New tab", null, panel, null);

        JPanel panel_1 = new JPanel();
        tabbedPane.addTab("New tab", null, panel_1, null);

        JPanel panel_2 = new JPanel();
        tabbedPane.addTab("New tab", null, panel_2, null);

        .. rest of components.

10个班级。

我正在尝试扩充图像并将其保存到磁盘，但我想保留文件夹结构，所以：

/howler-monkey/
    1.jpg
    2.jpg
    ...jpg
/japanese-mcaque
    1.jpg
    2.jpg
    ...

似乎我只是用

运行

/augmented/
    /howler-monkey
        aug_1.jpg
        aug_2.jpg
    /japanese-mcaque
        aug_1.jpg
        aug_2.jpg

它会将增强的图像转储到trainDataGenerator = ImageDataGenerator(shear_range=0.2, zoom_range=0.2, horizontal_flip=True, rotation_range=20, width_shift_range=0.2, height_shift_range=0.2, fill_mode='nearest') fileIterator = trainDataGenerator.flow_from_directory('{}/training'.format(args.dataset), save_to_dir='{}/{}'.format(args.dataset, args.output)) i = 0 for image in fileIterator: if i > 10: break文件夹中，但它不会保存目录结构，因此很难用于训练。

如何在扩充图像时保留原始目录结构？

Answer 1

所以我最终只是使用pathlib和trainDataGenerator = ImageDataGenerator(shear_range=0.2, zoom_range=0.2, horizontal_flip=True, rotation_range=20, width_shift_range=0.2, height_shift_range=0.2) for path in list_images(args.dataset): img = cv2.imread(path) img = img_to_array(img) img = np.expand_dims(img, axis=0) pathlib.Path('{}/{}/{}'.format(args.dataset, args.output, path.split(os.path.sep)[-2])).mkdir( parents=True, exist_ok=True) print(path) total = 0 for image in trainDataGenerator.flow(img, batch_size=1, save_to_dir='{}/{}/{}'.format(args.dataset, args.output, path.split(os.path.sep)[-2]), save_format='jpeg'): print(total) total += 1 if total == 10: break手动创建目录：

args.dataset

其中args.output是包含训练图像的str，而augmentedImages是包含declare module 'node-helper-lib' { import * as SomeThirdParty from 'node-helper-lib'; interface Helper { new(opt: SomeThirdParty.Options): SomeThirdParty.Type } export = Helper; }的str。

使用Keras＆＃39;时的文件夹结构ImageDataGenerator

1 个答案: