def text_dataset_from_directory(directory, labels='inferred', label_mode='int', class_names=None, batch_size=32, max_length=None, shuffle=True, seed=None, validation_split=None, subset=None, follow_links=False): """Generates a `tf.data.Dataset` from text files in a directory. If your directory structure is: ``` main_directory/ ...class_a/ ......a_text_1.txt ......a_text_2.txt ...class_b/ ......b_text_1.txt ......b_text_2.txt ``` Then calling `text_dataset_from_directory(main_directory, labels='inferred')` will return a `tf.data.Dataset` that yields batches of texts from the subdirectories `class_a` and `class_b`, together with labels 0 and 1 (0 corresponding to `class_a` and 1 corresponding to `class_b`). Only `.txt` files are supported at this time. Args: directory: Directory where the data is located. If `labels` is "inferred", it should contain subdirectories, each containing text files for a class. Otherwise, the directory structure is ignored. labels: Either "inferred" (labels are generated from the directory structure), None (no labels), or a list/tuple of integer labels of the same size as the number of text files found in the directory. Labels should be sorted according to the alphanumeric order of the text file paths (obtained via `os.walk(directory)` in Python). label_mode: - 'int': means that the labels are encoded as integers (e.g. for `sparse_categorical_crossentropy` loss). - 'categorical' means that the labels are encoded as a categorical vector (e.g. for `categorical_crossentropy` loss). - 'binary' means that the labels (there can be only 2) are encoded as `float32` scalars with values 0 or 1 (e.g. for `binary_crossentropy`). - None (no labels). class_names: Only valid if "labels" is "inferred". This is the explicit list of class names (must match names of subdirectories). Used to control the order of the classes (otherwise alphanumerical order is used). batch_size: Size of the batches of data. Default: 32. max_length: Maximum size of a text string. Texts longer than this will be truncated to `max_length`. shuffle: Whether to shuffle the data. Default: True. If set to False, sorts the data in alphanumeric order. seed: Optional random seed for shuffling and transformations. validation_split: Optional float between 0 and 1, fraction of data to reserve for validation. subset: One of "training" or "validation". Only used if `validation_split` is set. follow_links: Whether to visits subdirectories pointed to by symlinks. Defaults to False. Returns: A `tf.data.Dataset` object. - If `label_mode` is None, it yields `string` tensors of shape `(batch_size,)`, containing the contents of a batch of text files. - Otherwise, it yields a tuple `(texts, labels)`, where `texts` has shape `(batch_size,)` and `labels` follows the format described below. Rules regarding labels format: - if `label_mode` is `int`, the labels are an `int32` tensor of shape `(batch_size,)`. - if `label_mode` is `binary`, the labels are a `float32` tensor of 1s and 0s of shape `(batch_size, 1)`. - if `label_mode` is `categorial`, the labels are a `float32` tensor of shape `(batch_size, num_classes)`, representing a one-hot encoding of the class index. """ if labels not in ('inferred', None): if not isinstance(labels, (list, tuple)): raise ValueError( '`labels` argument should be a list/tuple of integer labels, of ' 'the same size as the number of text files in the target ' 'directory. If you wish to infer the labels from the subdirectory ' 'names in the target directory, pass `labels="inferred"`. ' 'If you wish to get a dataset that only contains text samples ' f'(no labels), pass `labels=None`. Received: labels={labels}') if class_names: raise ValueError('You can only pass `class_names` if ' f'`labels="inferred"`. Received: labels={labels}, and ' f'class_names={class_names}') if label_mode not in {'int', 'categorical', 'binary', None}: raise ValueError( '`label_mode` argument must be one of "int", "categorical", "binary", ' f'or None. Received: label_mode={label_mode}') if labels is None or label_mode is None: labels = None label_mode = None dataset_utils.check_validation_split_arg( validation_split, subset, shuffle, seed) if seed is None: seed = np.random.randint(1e6) file_paths, labels, class_names = dataset_utils.index_directory( directory, labels, formats=('.txt',), class_names=class_names, shuffle=shuffle, seed=seed, follow_links=follow_links) if label_mode == 'binary' and len(class_names) != 2: raise ValueError( f'When passing `label_mode="binary"`, there must be exactly 2 ' f'class_names. Received: class_names={class_names}') file_paths, labels = dataset_utils.get_training_or_validation_split( file_paths, labels, validation_split, subset) if not file_paths: raise ValueError(f'No text files found in directory {directory}. ' f'Allowed format: .txt') dataset = paths_and_labels_to_dataset( file_paths=file_paths, labels=labels, label_mode=label_mode, num_classes=len(class_names), max_length=max_length) if shuffle: # Shuffle locally at each iteration dataset = dataset.shuffle(buffer_size=batch_size * 8, seed=seed) dataset = dataset.prefetch(tf.data.AUTOTUNE).batch(batch_size) # Users may need to reference `class_names`. dataset.class_names = class_names return dataset
def text_dataset_from_directory( directory: str, batch_size: int = 32, max_length: Optional[int] = None, shuffle: bool = True, seed: Optional[int] = None, validation_split: Optional[float] = None, subset: Optional[str] = None, ) -> tf.data.Dataset: """Generates a `tf.data.Dataset` from text files in a directory. If your directory structure is: ``` main_directory/ ...class_a/ ......a_text_1.txt ......a_text_2.txt ...class_b/ ......b_text_1.txt ......b_text_2.txt ``` Then calling `text_dataset_from_directory(main_directory)` will return a `tf.data.Dataset` that yields batches of texts from the subdirectories `class_a` and `class_b`, together with labels 'class_a' and 'class_b'. Only `.txt` files are supported at this time. # Arguments directory: Directory where the data is located. If `labels` is "inferred", it should contain subdirectories, each containing text files for a class. Otherwise, the directory structure is ignored. batch_size: Size of the batches of data. Defaults to 32. max_length: Maximum size of a text string. Texts longer than this will be truncated to `max_length`. shuffle: Whether to shuffle the data. Default: True. If set to False, sorts the data in alphanumeric order. seed: Optional random seed for shuffling and transformations. validation_split: Optional float between 0 and 1, fraction of data to reserve for validation. subset: One of "training" or "validation". Only used if `validation_split` is set. # Returns A `tf.data.Dataset` object, which yields a tuple `(texts, labels)`, where both has shape `(batch_size,)` and type of tf.string. """ if seed is None: seed = np.random.randint(1e6) file_paths, labels, class_names = dataset_utils.index_directory( directory, "inferred", formats=(".txt", ), shuffle=shuffle, seed=seed) file_paths, labels = get_training_or_validation_split( file_paths, labels, validation_split, subset) strings = tf.data.Dataset.from_tensor_slices(file_paths) strings = strings.map(tf.io.read_file) if max_length is not None: strings = strings.map(lambda x: tf.strings.substr(x, 0, max_length)) labels = np.array(class_names)[np.array(labels)] labels = tf.data.Dataset.from_tensor_slices(labels) dataset = tf.data.Dataset.zip((strings, labels)) dataset = dataset.batch(batch_size) return dataset
def image_dataset_from_directory(directory, labels='inferred', label_mode='int', class_names=None, color_mode='rgb', batch_size=32, image_size=(256, 256), shuffle=True, seed=None, validation_split=None, subset=None, interpolation='bilinear', follow_links=False, smart_resize=False): """Generates a `tf.data.Dataset` from image files in a directory. If your directory structure is: ``` main_directory/ ...class_a/ ......a_image_1.jpg ......a_image_2.jpg ...class_b/ ......b_image_1.jpg ......b_image_2.jpg ``` Then calling `image_dataset_from_directory(main_directory, labels='inferred')` will return a `tf.data.Dataset` that yields batches of images from the subdirectories `class_a` and `class_b`, together with labels 0 and 1 (0 corresponding to `class_a` and 1 corresponding to `class_b`). Supported image formats: jpeg, png, bmp, gif. Animated gifs are truncated to the first frame. Args: directory: Directory where the data is located. If `labels` is "inferred", it should contain subdirectories, each containing images for a class. Otherwise, the directory structure is ignored. labels: Either "inferred" (labels are generated from the directory structure), None (no labels), or a list/tuple of integer labels of the same size as the number of image files found in the directory. Labels should be sorted according to the alphanumeric order of the image file paths (obtained via `os.walk(directory)` in Python). label_mode: - 'int': means that the labels are encoded as integers (e.g. for `sparse_categorical_crossentropy` loss). - 'categorical' means that the labels are encoded as a categorical vector (e.g. for `categorical_crossentropy` loss). - 'binary' means that the labels (there can be only 2) are encoded as `float32` scalars with values 0 or 1 (e.g. for `binary_crossentropy`). - None (no labels). class_names: Only valid if "labels" is "inferred". This is the explict list of class names (must match names of subdirectories). Used to control the order of the classes (otherwise alphanumerical order is used). color_mode: One of "grayscale", "rgb", "rgba". Default: "rgb". Whether the images will be converted to have 1, 3, or 4 channels. batch_size: Size of the batches of data. Default: 32. image_size: Size to resize images to after they are read from disk. Defaults to `(256, 256)`. Since the pipeline processes batches of images that must all have the same size, this must be provided. shuffle: Whether to shuffle the data. Default: True. If set to False, sorts the data in alphanumeric order. seed: Optional random seed for shuffling and transformations. validation_split: Optional float between 0 and 1, fraction of data to reserve for validation. subset: One of "training" or "validation". Only used if `validation_split` is set. interpolation: String, the interpolation method used when resizing images. Defaults to `bilinear`. Supports `bilinear`, `nearest`, `bicubic`, `area`, `lanczos3`, `lanczos5`, `gaussian`, `mitchellcubic`. follow_links: Whether to visits subdirectories pointed to by symlinks. Defaults to False. smart_resize: If True, the resizing function used will be `tf.keras.preprocessing.image.smart_resize`, which preserves the aspect ratio of the original image by using a mixture of resizing and cropping. If False (default), the resizing function is `tf.image.resize`, which does not preserve aspect ratio. Returns: A `tf.data.Dataset` object. - If `label_mode` is None, it yields `float32` tensors of shape `(batch_size, image_size[0], image_size[1], num_channels)`, encoding images (see below for rules regarding `num_channels`). - Otherwise, it yields a tuple `(images, labels)`, where `images` has shape `(batch_size, image_size[0], image_size[1], num_channels)`, and `labels` follows the format described below. Rules regarding labels format: - if `label_mode` is `int`, the labels are an `int32` tensor of shape `(batch_size,)`. - if `label_mode` is `binary`, the labels are a `float32` tensor of 1s and 0s of shape `(batch_size, 1)`. - if `label_mode` is `categorial`, the labels are a `float32` tensor of shape `(batch_size, num_classes)`, representing a one-hot encoding of the class index. Rules regarding number of channels in the yielded images: - if `color_mode` is `grayscale`, there's 1 channel in the image tensors. - if `color_mode` is `rgb`, there are 3 channel in the image tensors. - if `color_mode` is `rgba`, there are 4 channel in the image tensors. """ if labels not in ('inferred', None): if not isinstance(labels, (list, tuple)): raise ValueError( '`labels` argument should be a list/tuple of integer labels, of ' 'the same size as the number of image files in the target ' 'directory. If you wish to infer the labels from the subdirectory ' 'names in the target directory, pass `labels="inferred"`. ' 'If you wish to get a dataset that only contains images ' '(no labels), pass `label_mode=None`.') if class_names: raise ValueError( 'You can only pass `class_names` if the labels are ' 'inferred from the subdirectory names in the target ' 'directory (`labels="inferred"`).') if label_mode not in {'int', 'categorical', 'binary', None}: raise ValueError( '`label_mode` argument must be one of "int", "categorical", "binary", ' 'or None. Received: %s' % (label_mode, )) if labels is None or label_mode is None: labels = None label_mode = None if color_mode == 'rgb': num_channels = 3 elif color_mode == 'rgba': num_channels = 4 elif color_mode == 'grayscale': num_channels = 1 else: raise ValueError( '`color_mode` must be one of {"rbg", "rgba", "grayscale"}. ' 'Received: %s' % (color_mode, )) interpolation = image_preprocessing.get_interpolation(interpolation) dataset_utils.check_validation_split_arg(validation_split, subset, shuffle, seed) if seed is None: seed = np.random.randint(1e6) image_paths, labels, class_names = dataset_utils.index_directory( directory, labels, formats=ALLOWLIST_FORMATS, class_names=class_names, shuffle=shuffle, seed=seed, follow_links=follow_links) if label_mode == 'binary' and len(class_names) != 2: raise ValueError( 'When passing `label_mode="binary", there must exactly 2 classes. ' 'Found the following classes: %s' % (class_names, )) image_paths, labels = dataset_utils.get_training_or_validation_split( image_paths, labels, validation_split, subset) if not image_paths: raise ValueError('No images found.') dataset = paths_and_labels_to_dataset(image_paths=image_paths, image_size=image_size, num_channels=num_channels, labels=labels, label_mode=label_mode, num_classes=len(class_names), interpolation=interpolation, smart_resize=smart_resize) if shuffle: # Shuffle locally at each iteration dataset = dataset.shuffle(buffer_size=batch_size * 8, seed=seed) dataset = dataset.batch(batch_size) # Users may need to reference `class_names`. dataset.class_names = class_names # Include file paths for images as attribute. dataset.file_paths = image_paths return dataset
def image_dataset_from_directory( directory: str, batch_size: int = 32, color_mode: str = "rgb", image_size: Tuple[int, int] = (256, 256), interpolation: str = "bilinear", shuffle: bool = True, seed: Optional[int] = None, validation_split: Optional[float] = None, subset: Optional[str] = None, ) -> tf.data.Dataset: """Generates a `tf.data.Dataset` from image files in a directory. If your directory structure is: ``` main_directory/ ...class_a/ ......a_image_1.jpg ......a_image_2.jpg ...class_b/ ......b_image_1.jpg ......b_image_2.jpg ``` Then calling `image_dataset_from_directory(main_directory)` will return a `tf.data.Dataset` that yields batches of images from the subdirectories `class_a` and `class_b`, together with labels 'class_a' and 'class_b'. Supported image formats: jpeg, png, bmp, gif. Animated gifs are truncated to the first frame. # Arguments directory: Directory where the data is located. If `labels` is "inferred", it should contain subdirectories, each containing images for a class. Otherwise, the directory structure is ignored. batch_size: Size of the batches of data. Default: 32. color_mode: One of "grayscale", "rgb", "rgba". Default: "rgb". Whether the images will be converted to have 1, 3, or 4 channels. image_size: Size to resize images to after they are read from disk. Defaults to `(256, 256)`. Since the pipeline processes batches of images that must all have the same size, this must be provided. interpolation: String, the interpolation method used when resizing images. Defaults to `bilinear`. Supports `bilinear`, `nearest`, `bicubic`, `area`, `lanczos3`, `lanczos5`, `gaussian`, `mitchellcubic`. shuffle: Whether to shuffle the data. Default: True. If set to False, sorts the data in alphanumeric order. seed: Optional random seed for shuffling and transformations. validation_split: Optional float between 0 and 1, fraction of data to reserve for validation. subset: One of "training" or "validation". Only used if `validation_split` is set. # Returns A `tf.data.Dataset` object, which yields a tuple `(texts, labels)`, where `images` has shape `(batch_size, image_size[0], image_size[1], num_channels)` where `labels` has shape `(batch_size,)` and type of tf.string. - if `color_mode` is `grayscale`, there's 1 channel in the image tensors. - if `color_mode` is `rgb`, there are 3 channel in the image tensors. - if `color_mode` is `rgba`, there are 4 channel in the image tensors. """ if color_mode == "rgb": num_channels = 3 elif color_mode == "rgba": num_channels = 4 elif color_mode == "grayscale": num_channels = 1 else: raise ValueError( '`color_mode` must be one of {"rbg", "rgba", "grayscale"}. ' "Received: %s" % (color_mode, )) if seed is None: seed = np.random.randint(1e6) image_paths, labels, class_names = dataset_utils.index_directory( directory, "inferred", formats=WHITELIST_FORMATS, shuffle=shuffle, seed=seed) image_paths, labels = get_training_or_validation_split( image_paths, labels, validation_split, subset) images = tf.data.Dataset.from_tensor_slices(image_paths) images = images.map(lambda img: path_to_image(img, num_channels, image_size, interpolation)) labels = np.array(class_names)[np.array(labels)] labels = tf.data.Dataset.from_tensor_slices(labels) dataset = tf.data.Dataset.zip((images, labels)) dataset = dataset.batch(batch_size) return dataset