如何为Keras中的DataGenerator选择图像进行预测？

Question

I load the images in the following format for bulk predictions. This method works fine if all images are in a parent folder as follows :

..Parent Folder > Sub folder > All image files.

So I need to set my 'directory' = 'Parent Folder'. It reads all the sub folders inside 'Parent folder' and their images.

predict_datagen = ImageDataGenerator(rescale = 1./255)
            predict_generator = predict_datagen.flow_from_directory(directory = parent_image_folder,
                                                target_size = (size, size),
                                                batch_size = batch_size,
                                                class_mode = None, 
                                                shuffle = False)

predict_generator.reset()
            predict_filenames = predict_generator.filenames
            predictions = model.predict_generator(predict_generator, verbose=1, steps = int(len(predict_generator.filenames) / batch_size)+1)

But in my project, I need to load only the images in selected subfolders folders in 'Parent folder'. I could achieve it by copying the selected folders into a separate folder(new parent folder). But is there any other better method?

Answer 1

There is no way to do this without modifying ImageDataGenerator. If you're on linux, the best way I can think of is to create a separate parent directory, and then create symlinks for the selected subfolders. Make sure to set follow_links=True in flow_from_directory .