Face Trigger - Utils package

Contains modules for performing common utilities.

Submodules

face_trigger.utils.common module

Module that contains various common utilities.

class face_trigger.utils.common.RepeatedTimer(interval, function, *args, **kwargs)

Bases: object

Creates a timer object that can call a specified function repeatedly after a specified delayed. Does not drift over time.

start()

Starts the timer thread and schedules the next call

stop()

Stops the timer thread

face_trigger.utils.common.clamp_rectangle(x1=None, y1=None, x2=None, y2=None, x1_min=0, y1_min=0, x2_max=None, y2_max=None)

Clamps the coordinates of a rectangle to within specified limits

Parameters:
  • x1 (int) – the leftmost x-coordinate
  • y1 (int) – the topmost y-coordinate
  • x2 (int) – the rightmost x-coordinate
  • y2 (int) – the bottommost y-coordinate
  • x1_min (int) – the leftmost possible x-coordinate
  • y1 – the topmost possible y-coordinate
  • x2 – the rightmost possible x-coordinate
  • y2 – the bottommost possible y-coordinate
Returns:

clamped coordinates (x1_clamped, y1_clamped, x2_clamped, y2_clamped)
Return type:

4-tuple
face_trigger.utils.common.shape_to_np(shape, dtype='int')

Convert a dlib.shape to a numpy array

Parameters:shape (dlib.shape) – a dlib.shape object
Returns:converted numpy array
Return type:numpy.array

face_trigger.utils.data module

Module that contains functions related to dataset cleaning.

class face_trigger.utils.data.Dataset(dataset_path=None, split_path=None)

A class for generating manual train-test splits and loading data from generated splits.

>>> dataset = Dataset(dataset_path="/media/ankurrc/new_volume/softura/facerec/datasets/norm_standard_att")
>>> folds = 3
>>> training_samples = [2, 5, 8]
>>> dataset.split(num_train_list=training_samples, folds=folds,
                split_path="/media/ankurrc/new_volume/softura/facerec/split_path")
load_data(split_path=None, is_train=None, num_train=None, fold=None)

Gets the test or train data

Parameters:
  • is_train (bool) – a flag to indicate whether we need training data or test data
  • fold (int) – the fold for which data needs to be loaded
  • num_train (int) – the subdirectory indicating number of training samples per subject
  • split_path (str) – path to store the dataset train-test split information. Will rename the old directory, if it exists.
Returns:

a tuple of a vector of faces and it’s corresponding labels
Return type:

(numpy.ndarray, numpy.ndarray)
split(split_path=None, num_train_list=None, folds=1)

Generates a train-test split based on the number of training samples

Parameters:
  • num_train_list (list of inttegers) – number of training samples per fold
  • fold (int) – total number of folds
  • split_path (str) – path to store the dataset train-test split information. Will rename the old directory, if it exists.
face_trigger.utils.data.dataset_filter(dataset_path=None, output_path=None)

Filter the dataset by: 1. A face is detected in the image. If no face or more than one face is detected, the image is rejected. 2. For each detected face, 5-landmarks are detected. If landmarks are not detected, image is rejected. 3. A new dataset is created by not including the rejected images.

Parameters:
  • dataset_path (str) – path to the original dataset
  • output_path (str) – path to the filtered dataset
Returns:

dictionary of rejected images
Return type:

dict

 
>>> dataset_path = "/media/ankurrc/new_volume/softura/facerec/datasets/standard_att_copy"
>>> output_path = "/media/ankurrc/new_volume/softura/facerec/att_norm"

>>> dataset_filter(
    dataset_path=dataset_path, output_path=output_path)
face_trigger.utils.data.get_jittered_images(image_path, num_jitters=5, disturb_colors=False)

face_trigger.utils.train module

Module that contains functions for loading embeddings from dataset during training.

face_trigger.utils.train.generate_embeddings_for_dataset(dataset_path=None)

Generates embeddings by sequentially reading images from the dataset.

Parameters:dataset_path – path to the dataset
Returns:X, y where X is a list of numpy array (128-d vectors) and y is a numpy array representing the labels
face_trigger.utils.train.generate_embeddings_for_split_and_fold(dataset_path=None, split_path=None, fold=None, num_train=None)

Generate embeddings of dataset for a particular split and fold.

Parameters:
  • dataset_path (str) – path to the dataset
  • split_path (str) – path to the directory holding train-test split info
  • fold (int) – which fold to generate embeddings on
  • num_train (int) – folder name, indacating the number of training samples per subject
Returns:

X_train, y_train, X_test, y_test

Module contents