Functions
def	_parse_labels (str label_path)

def	_normalize_labels (label, int n, int l)

tf.data.Dataset	_get_path_dataset (str path, int seed, bool shuffle=True)

def	_get_padded_labels (data, r, R, height, width)

tuple[tf.Tensor, tf.Tensor]	_preprocess_function (tf.Tensor data_x, tf.Tensor data_y, tuple[int] image_size, str interpolation, str aspect_ratio, str color_mode, int nbr_keypoints)

Tuple[tf.data.Dataset, tf.data.Dataset]	_get_train_val_ds (str training_path, tuple[int] image_size=None, int nbr_keypoints=None, str interpolation=None, str aspect_ratio=None, str color_mode=None, float validation_split=None, int batch_size=None, int seed=None, bool shuffle=True, bool to_cache=False)

tf.data.Dataset	_get_ds (str data_path=None, tuple[int] image_size=None, int nbr_keypoints=None, str interpolation=None, str aspect_ratio=None, str color_mode=None, int batch_size=None, int seed=None, bool shuffle=False, bool to_cache=False)

Tuple[tf.data.Dataset, tf.data.Dataset, tf.data.Dataset]	load_dataset (str dataset_name=None, str training_path=None, str validation_path=None, str quantization_path=None, str test_path=None, float validation_split=None, int nbr_keypoints=None, tuple[int] image_size=None, str interpolation=None, str aspect_ratio=None, str color_mode=None, int batch_size=None, int seed=None)

Function Documentation

◆ _get_ds()

tf.data.Dataset data_loader._get_ds	(	str	data_path = `None`,
		tuple[int]	image_size = `None`,
		int	nbr_keypoints = `None`,
		str	interpolation = `None`,
		str	aspect_ratio = `None`,
		str	color_mode = `None`,
		int	batch_size = `None`,
		int	seed = `None`,
		bool	shuffle = `False`,
		bool	to_cache = `False`
	)

private

Loads the images from the given dataset root directory and returns a tf.data.Dataset.
The dataset has the following directory structure (checked in parse_config.py):
    dataset_root_dir:
        image_1.jpg
        image_1.txt
        ...
        image_2.jpg
        image_2.txt

Args:
    data_path (str): Path to the directory containing the images.
    image_size (tuple[int]): Size of the input images to resize them to.
    nbr_keypoints (int): number of keypoints for a person
    interpolation (str): Interpolation method to use when resizing the images.
    aspect_ratio (bool): Whether or not to crop the images to the specified aspect ratio.
    color_mode (str): Color mode to use for the images.
    batch_size (int): Batch size to use for the dataset.
    seed (int): Seed to use for shuffling the data.
    shuffle (bool): Whether or not to shuffle the data.
    to_cache (bool): Whether or not to cache the dataset.

Returns:
    tf.data.Dataset: Dataset containing the images.

Definition at line 284 of file data_loader.py.

References _get_path_dataset(), and _preprocess_function().

Referenced by load_dataset().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ _get_padded_labels()

def data_loader._get_padded_labels	(	data,
		r,
		R,
		height,
		width
	)

private

Definition at line 116 of file data_loader.py.

Referenced by _preprocess_function().

Here is the caller graph for this function:

◆ _get_path_dataset()

tf.data.Dataset data_loader._get_path_dataset	(	str	path,
		int	seed,
		bool	shuffle = `True`
	)

private

Creates a tf.data.Dataset from a dataset root directory path.
The dataset has the following directory structure (checked in parse_config.py):
    dataset_root_dir:
        image_1.jpg
        image_1.txt
        ...
        image_2.jpg
        image_2.txt

Args:
    path (str): Path of the dataset folder.
    seed (int): seed when performing shuffle.
    shuffle (bool): Shuffle the dataset.

Returns:
    dataset(tf.data.Dataset) -> dataset with a tuple (path, label) of each sample.

Definition at line 67 of file data_loader.py.

References _normalize_labels().

Referenced by _get_ds(), and _get_train_val_ds().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ _get_train_val_ds()

Tuple[tf.data.Dataset, tf.data.Dataset] data_loader._get_train_val_ds	(	str	training_path,
		tuple[int]	image_size = `None`,
		int	nbr_keypoints = `None`,
		str	interpolation = `None`,
		str	aspect_ratio = `None`,
		str	color_mode = `None`,
		float	validation_split = `None`,
		int	batch_size = `None`,
		int	seed = `None`,
		bool	shuffle = `True`,
		bool	to_cache = `False`
	)

private

Loads the images under a given dataset root directory and returns training 
and validation tf.Data.datasets.
The dataset has the following directory structure (checked in parse_config.py):
    dataset_root_dir:
        image_1.jpg
        image_1.txt
        ...
        image_2.jpg
        image_2.txt

Args:
    training_path (str): Path to the directory containing the training images.
    image_size (tuple[int]): Size of the input images to resize them to.
    nbr_keypoints (int): number of keypoints for a person
    interpolation (float): Interpolation method to use when resizing the images.
    aspect_ratio (bool): Whether or not to crop the images to the specified aspect ratio.
    color_mode (str): Color mode to use for the images.
    validation_split (float): Fraction of the data to use for validation.
    batch_size (int): Batch size to use for training and validation.
    seed (int): Seed to use for shuffling the data.
    shuffle (bool): Whether or not to shuffle the data.
    to_cache (bool): Whether or not to cache the datasets.

Returns:
    Tuple[tf.data.Dataset, tf.data.Dataset]: Training and validation datasets.

Definition at line 203 of file data_loader.py.

References _get_path_dataset(), and _preprocess_function().

Referenced by load_dataset().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ _normalize_labels()

def data_loader._normalize_labels	(		label,
		int	n,
		int	l
	)

private

Normalization of the labels -> same shape for every label regarding the number of ground truths
Args:
    label (np.array): shape (ground_truths, 5+3*keypoints) ground truths present in the label file
    l     (int): shape (1, ) maximum number of ground truths present in a label file
    n     (int): shape (1, ) current number of ground truths present in this label file

Returns:
    normalized_label (np.array) : shape (l, 5+3*keypoints) label with normalized shape

Definition at line 51 of file data_loader.py.

Referenced by _get_path_dataset().

Here is the caller graph for this function:

◆ _parse_labels()

def data_loader._parse_labels ( str label_path )

private

Parsing of the labels files
Args:
    label_path (str): Path of the label file.

Returns:
    ground_truths (np.array) : shape (ground_truths, 5+3*keypoints) ground truths present in the label file

Definition at line 32 of file data_loader.py.

◆ _preprocess_function()

tuple[tf.Tensor, tf.Tensor] data_loader._preprocess_function	(	tf.Tensor	data_x,
		tf.Tensor	data_y,
		tuple[int]	image_size,
		str	interpolation,
		str	aspect_ratio,
		str	color_mode,
		int	nbr_keypoints
	)

private

Load images from path and apply necessary transformations.

Definition at line 171 of file data_loader.py.

References _get_padded_labels().

Referenced by _get_ds(), and _get_train_val_ds().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ load_dataset()

Tuple[tf.data.Dataset, tf.data.Dataset, tf.data.Dataset] data_loader.load_dataset	(	str	dataset_name = `None`,
		str	training_path = `None`,
		str	validation_path = `None`,
		str	quantization_path = `None`,
		str	test_path = `None`,
		float	validation_split = `None`,
		int	nbr_keypoints = `None`,
		tuple[int]	image_size = `None`,
		str	interpolation = `None`,
		str	aspect_ratio = `None`,
		str	color_mode = `None`,
		int	batch_size = `None`,
		int	seed = `None`
	)

Loads the images from the given dataset root directories and returns training,
validation, and test tf.data.Datasets.
The datasets have the following directory structure (checked in parse_config.py):
    dataset_root_dir:
        image_1.jpg
        image_1.txt
        ...
        image_2.jpg
        image_2.txt

Args:
    dataset_name (str): Name of the dataset to load.
    training_path (str): Path to the directory containing the training images.
    validation_path (str): Path to the directory containing the validation images.
    quantization_path (str): Path to the directory containing the quantization images.
    test_path (str): Path to the directory containing the test images.
    validation_split (float): Fraction of the data to use for validation.
    nbr_keypoints (int): number of keypoints for a person
    image_size (tuple[int]): resizing (width, height) of input images
    interpolation (str): Interpolation method to use when resizing the images.
    aspect_ratio (bool): Whether or not to crop the images to the specified aspect ratio.
    color_mode (str): Color mode to use for the images.
    batch_size (int): Batch size to use for the datasets.
    seed (int): Seed to use for shuffling the data.

Returns:
    Tuple[tf.data.Dataset, tf.data.Dataset, tf.data.Dataset]: Training, validation, and test datasets.

Definition at line 352 of file data_loader.py.

References _get_ds(), and _get_train_val_ds().

Referenced by preprocess.preprocess().

Here is the call graph for this function:

Here is the caller graph for this function:

Functions

Function Documentation

◆ _get_ds()

◆ _get_padded_labels()

◆ _get_path_dataset()

◆ _get_train_val_ds()

◆ _normalize_labels()

◆ _parse_labels()

◆ _preprocess_function()

◆ load_dataset()