wound_segmentation.utils module

Utility functions for wound segmentation evaluation and result visualization.

This module contains helper functions used throughout the segmentation pipeline, including:

  • Metric computation (e.g., Dice coefficient and Dice-based loss)

  • Combined loss function (BCE + Dice) for model training.

  • Visualization and saving of segmentation results.

These utilities are designed to be modular and testable.

Functions

  • dice_coef : Compute Dice similarity coefficient between two binary masks.

  • dice_loss : Calculate Dice loss from Dice coefficient.

  • bce_dice_loss_weighted : Weighted combination of BCE and Dice losses.

  • save_result : Generate and save a 3-panel plot (input, mask, overlay).

wound_segmentation.utils.dice_coef(y_true: tensorflow.Tensor, y_pred: tensorflow.Tensor, smooth: float = 1e-06) → tensorflow.Tensor

This function computes the Dice coefficient, a measure of similarity between two binary masks.

Parameters:

  • y_true (tf.Tensor) -- Ground truth binary mask.

  • y_pred (tf.Tensor) -- Predicted binary mask.

  • smooth (float, optional) -- Smoothing constant to avoid division by zero (default is 1e-6).

Returns:

Dice similarity coefficient.

Return type:

tf.Tensor

Raises:

TypeError -- If y_true or y_pred is not a tf.Tensor.

Examples

>>> dice_coef(y_true, y_pred)
wound_segmentation.utils.dice_loss(y_true: tensorflow.Tensor, y_pred: tensorflow.Tensor) → tensorflow.Tensor

This function computes the Dice loss, defined as 1 - Dice coefficient.

Parameters:

  • y_true (tf.Tensor) -- Ground truth binary mask.

  • y_pred (tf.Tensor) -- Predicted binary mask.

Returns:

Dice loss value in the range [0, 1].

Return type:

tf.Tensor

Examples

>>> dice = dice_loss(y_true, y_pred)
wound_segmentation.utils.bce_dice_loss_weighted(y_true: tensorflow.Tensor, y_pred: tensorflow.Tensor, alpha: float = 0.2) → tensorflow.Tensor

This function computes a weighted combination of Binary Crossentropy and Dice loss.

Parameters:

  • y_true (tf.Tensor) -- Ground truth binary mask.

  • y_pred (tf.Tensor) -- Predicted binary mask..

  • alpha (float, optional) -- Weight for BCE loss; (1 - alpha) is used for Dice loss.

Returns:

Weighted loss value.

Return type:

tf.Tensor

Raises:

ValueError -- If alpha is not in the range [0, 1].

Examples

>>> bce_dice_loss_weighted(y_true, y_pred, alpha=0.3)
wound_segmentation.utils.save_result(image: numpy.ndarray, mask: numpy.ndarray, basename: str, output_dir: str) → None

This function saves the input image, predicted mask and the overlay of the mask on the image, side-by-side as a PNG.

Parameters:

  • image (np.ndarray) -- Input RGB image of shape (H, W, 3).

  • mask (np.ndarray) -- Binary predicted mask.

  • basename (str) -- Output image file name prefix.

  • output_dir (str) -- Directory path to save the result image.

Raises:

OSError -- If the output directory cannot be created.

Examples

>>> save_result(image, mask, "sample", "outputs/")