image_handler.py 文件源码

def read_data(self, train_split=0.80, dev_split=0.10, test_split=0.10):
        """
        Class function to read images from `self.image_dir`and split them into three groups: train/dev/test.
        """
        assert (train_split + dev_split + test_split == 1.0)

        all_images = glob.glob(self.image_dir + "*.png")
        data = []

        for image_path in all_images:
            image = imread(image_path, flatten=True)
            image = image.reshape(IMAGE_WIDTH*IMAGE_HEIGHT)
            # image = np.multiply(image, 1.0 / 255.0) No scaling here

            data.append(image)

        data = np.array(data)
        data = data.astype(np.uint8)

        total_images = data.shape[0]

        train_limit = int(total_images * train_split)
        dev_limit = train_limit + int(total_images * dev_split)

        self.train = data[:train_limit]
        self.dev = data[train_limit:dev_limit]
        self.test = data[dev_limit:]

        # Only shuffling training data.
        random.shuffle(self.train)

        self.data_dict = {
            'train':    self.train,
            'dev':      self.dev,
            'test':     self.test}