Helper functions for Training of GlottisNetV2

Some helper functions for the training are contained in Utils/data.py

• load_data(aplist, n):

  Load anterior and posterior points to Dataframe.

  • Parameters:

    • aplist (JSON-file): Contains anterior and posterior points of each image (output of annotation tool)

    • n (int): number of training images

  • Returns:

    • DataFrame with three columns: image id, coordinates of anterior, and coordinates of posterior points.

• img_moment(img):

  Calculate mass of input image

  • Parameters:

    • img: Prediction map of anterior and posterior point

  • Returns:

    • Coordinates of mass of input image

• MAPE(keypoints_orig, keypoints_pred, nr_points, eps = 1e-9):

  Calculate MAPE

  • Parameters:

    • keypoints_orig: Coordinates of original keypoints

    • keypoints_pred: Coordinates of predicted keypoints

    • nr_points: number of coordinates (in this case 4, two for anterior and posterior points each)

  • Returns:

    • MAPE

• MAPE3D(keypoints_orig, keypoints_pred, nr_points, eps = 1e-9):

  Calculate MAPE of several frames

  • Parameters:

    • keypoints_orig: Coordinates of original keypoints –> 5-dimensional array or 4-dimensional array with more than 3 channels

    • keypoints_pred: Coordinates of predicted keypoints –> 5-dimensional array or 4-dimensional array with more than 3 channels

    • nr_points: number of coordinates (in this case 4, two for anterior and posterior points each)

  • Returns:

    • MAPE

• metric_mape3D(y_true, y_pred):

  Metric to evaluate MAPE of several frames (custom metric). The input images have 5 dimensions or 4 dimensions and more than 2 channels.

  • Parameters

    • y_true: Coordinates of original keypoints –> (1,4)

    • y_pred: Coordinates of predicted keypoints –> (1,4)

  • Returns:

    • MAPE

• metric_mape(y_true, y_pred):

  Metric to evaluate MAPE (custom metric)

  • Parameters:

    • y_true: Coordinates of original keypoints –> (1,4)

    • y_pred: Coordinates of predicted keypoints –> (1,4)

  • Returns:

    • MAPE

• mape_ap(y_true, y_pred):

  Metric to evaluate MAPE of anterior point (custom metric)

  • Parameters:

    • y_true: Coordinates of original keypoints –> (1,2)

    • y_pred: Coordinates of predicted keypoints –> (1,2)

  • Returns:

    • MAPE of anterior point

• mape_ap(y_true, y_pred):

  Metric to evaluate MAPE of posterior point (custom meric)

  • Parameters:

    • y_true: Coordinates of original keypoints –> (1,2)

    • y_pred: Coordinates of predicted keypoints –> (1,2)

  • Returns:

    • MAPE of posterior point

• load_video_ap(aplist, nr):

  Used in TrainingChannles.ipynb, Training3DConv.ipynb and Training_LSTM.ipynb. Load anterior and posterior points to dictionary.

  • Parameters:

    • aplist: JSON-file

      • Contains anterior and posterior points of each image.

    • nr: int

      • video id

  • Returns:

    • Dictionary with three columns: video id, coordinates of anterior, and coordinates of posterior points.

• load_video_2D(aplist, nr, id_nr):

  Used in Training2DComp.ipynb and Training_3DGlottisNetV1. Load anterior and posterior points to dictionary.

  • Parameters

    • aplist: JSON-file

      • Contains anterior and posterior points of each image.

    • id_nr: int

      • video id

  • Returns:

    • Dictionary with three columns: video id, coordinates of anterior and coordinates of posterior points.

• MAPE_V1(keypoints_orig, keypoints_pred):

  Metric to evaluate MAPE with coordinates as input (GlottisNetV1)

  • Parameters:

    • y_true: Coordinates of original keypoints –> (1,4)

    • y_pred: Coordinates of predicted keypoints –> (1,4)

  • Returns:

    • MAPE

• MAPE_apV1(keypoints_orig, keypoints_pred):

  Metric to evaluate MAPE with coordinates as input (GlottisNetV1) for anterior point

  • Parameters:

    • y_true: Coordinates of original keypoints –> (1,2)

    • y_pred: Coordinates of predicted keypoints –> (1,2)

  • Returns:

    • MAPE for anterior point

• MAPE_ppV1(keypoints_orig, keypoints_pred):

  Metric to evaluate MAPE with coordinates as input (GlottisNetV1) for posterior point

  • Parameters:

    • y_true: Coordinates of original keypoints –> (1,2)

    • y_pred: Coordinates of predicted keypoints –> (1,2)

  • Returns:

    • MAPE for posterior point