tf-flower-dataset.py

import tensorflow as tf
AUTOTUNE = tf.data.experimental.AUTOTUNE

import pathlib
import random
import os
import IPython.display as display
import matplotlib.pyplot as plt


class FlowerDataset:
    def __init__(self):
        
        self.data_root = self.download_data()
        self.all_image_paths = self.get_img_paths()    
        self.label_names, self.all_image_labels = self.get_img_labels()
        
        #self.show_example()
        # 
        image_path = self.all_image_paths[0]
        label = self.all_image_labels[0]

        self.show_img(image_path, label)
        print()
        

    @staticmethod
    def preprocess_image(image):
        image = tf.image.decode_jpeg(image, channels=3)
        image = tf.image.resize(image, [192, 192])
        image /= 255.0  # normalize to [0,1] range

        return image

    
    def load_and_preprocess_image(self, path):
        image = tf.io.read_file(path)
        return self.preprocess_image(image)
    
    
    def show_img(self, img_path, label):
        plt.imshow(self.load_and_preprocess_image(img_path))
        plt.grid(False)
        plt.xlabel(self.caption_image(img_path))
        plt.title(self.label_names[label].title())
    
    
    
    # ---
    #  method
    # ---
    @staticmethod
    def download_data():
        data_root_orig = tf.keras.utils.get_file(origin='https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz',
                                         fname='flower_photos', untar=True)
        data_root = pathlib.Path(data_root_orig) # /root/.keras/datasets/flower_photos

        # show dirs
        for item in data_root.iterdir():
            print(item)
            
        return data_root
    
    
    def get_img_paths(self):
        all_image_paths = list(self.data_root.glob('*/*'))
        all_image_paths = [str(path) for path in all_image_paths]
        random.shuffle(all_image_paths)

        image_count = len(all_image_paths)
        print(image_count)

        all_image_paths[:10]
        return all_image_paths
    
    
    def get_img_labels(self):
        label_names = sorted(item.name for item in self.data_root.glob('*/') if item.is_dir())
        # ['daisy', 'dandelion', 'roses', 'sunflowers', 'tulips']

        label_to_index = dict((name, index) for index,name in enumerate(label_names))
        # {'daisy': 0, 'dandelion': 1, 'roses': 2, 'sunflowers': 3, 'tulips': 4}
        
        all_image_labels = [label_to_index[pathlib.Path(path).parent.name]
                            for path in self.all_image_paths]

        print("First 10 labels indices: ", all_image_labels[:10])
        return label_names, all_image_labels
    
        
    def show_example(self):
        for n in range(3):
            image_path = random.choice(self.all_image_paths)
            display.display(display.Image(image_path))
            print(self.caption_image(image_path))
            print()

        for n in range(3):
            image_path = random.choice(self.all_image_paths)
            display.display(display.Image(image_path))
            print(self.caption_image(image_path))
            print()
            

    def caption_image(self, image_path):
        attributions = (self.data_root/"LICENSE.txt").open(encoding='utf-8').readlines()[4:]
        attributions = [line.split(' CC-BY') for line in attributions]
        attributions = dict(attributions)
        
        image_rel = pathlib.Path(image_path).relative_to(self.data_root)
        return "Image (CC BY 2.0) " + ' - '.join(attributions[str(image_rel)].split(' - ')[:-1])
        
        
FlowerDataset()