liom_toolkit.segmentation.vseg.utils module

liom_toolkit.segmentation.vseg.utils.add_patch_to_empty_array(inference: ndarray, pred_y: ndarray, coords: tuple[int, int], stride: int, overlap: int, size: tuple[int, int]) → ndarray

Add a inferred patch to empty array

Parameters:

  • inference (np.ndarray) – The empty array to add the patch to

  • pred_y (np.ndarray) – The predicted patch

  • coords (tuple[int, int]) – The coordinates of the patch

  • stride (int) – The stride of the patch

  • overlap (int) – The overlap of the patch

  • size (tuple[int, int]) – The size of the patch

Returns:

The array with the patch added

Return type:

np.ndarray

liom_toolkit.segmentation.vseg.utils.calculate_metrics(y_true: ndarray, y_pred: ndarray) → list[float]

Calculate metrics between ground truth and prediction. Metrics are F1, Recall, Precision, Accuracy, and Jaccard.

Parameters:

  • y_true (np.ndarray) – Ground truth

  • y_pred (np.ndarray) – Prediction

Returns:

List of metrics

Return type:

list[float]

liom_toolkit.segmentation.vseg.utils.create_dir(path: str) → None

Create a directory if it does not exist yet

Parameters:

path (str) – The path to create

Returns:

None

liom_toolkit.segmentation.vseg.utils.create_patches(image_path: str, size: tuple[int, int] = (256, 256), stride: int = 64, norm: bool = False) → tuple[list[Any], tuple[int, ...], tuple[int, ...], ndarray[Any, dtype[Any]] | Any]

Create patches from an image

Parameters:

  • image_path (str) – The path to the image

  • size (tuple[int, int]) – The size of the patches

  • stride (int) – The stride of the patches

  • norm (bool) – Normalize the patches

Returns:

The patches, the shape of the image, the shape of the patches, and the image

Return type:

tuple[np.ndarray, tuple[int, int], tuple[int, int], np.ndarray]

liom_toolkit.segmentation.vseg.utils.crop_image(image: ndarray, size: tuple[int, int], stride: int) → ndarray

Crop an image to a specific size and stride

Parameters:

  • image (np.ndarray) – The image to crop

  • size (tuple[int, int]) – The size to crop to

  • stride (int) – The stride to crop with

Returns:

The cropped image

Return type:

np.ndarray

liom_toolkit.segmentation.vseg.utils.epoch_time(start_time: float, end_time: float) → tuple[int, int]

Calculate the elapsed time between start and end time

Parameters:

  • start_time (float) – The start time

  • end_time (float) – The end time

Returns:

The elapsed time in minutes and seconds

Return type:

tuple[int, int]

liom_toolkit.segmentation.vseg.utils.matplotlib_imshow(img: Tensor, one_channel: bool = False) → None

Visualize an image using matplotlib

Parameters:

  • img (torch.Tensor) – The image to visualize

  • one_channel (bool) – Whether the image has one channel

Returns:

None

liom_toolkit.segmentation.vseg.utils.numeric_filesort(path: str, folder: str = 'images', extension: str = 'png') → list[str]

Sort a list of filenames by numerical order

Parameters:

  • path (str) – The path to the folder

  • folder (str) – The folder to sort

  • extension (str) – The extension of the files

Returns:

The sorted list of filenames

Return type:

list[str]

liom_toolkit.segmentation.vseg.utils.patch(image_path: str, save_path: str, norm: bool, size: tuple[int, int] = (256, 256), stride: int = 64, augment: bool = True, threshold: int = 5, save_image: bool = True, use_mask: bool = True, remove_background_tiles: bool = False) → tuple[tuple[int, ...], tuple[int, ...], Any, Any] | tuple[tuple[int, ...], tuple[int, ...], Any]

Patch an image

Parameters:

  • image_path (str) – The path to the image

  • save_path (str) – The path to save the patches

  • norm (bool) – Normalize the image

  • size (tuple[int, int]) – The size of the patches

  • stride (int) – The stride of the patches

  • augment (bool) – Augment the patches

  • threshold (int) – The threshold for removing background tiles

  • save_image (bool) – Save the image

  • use_mask (bool) – Use a mask

  • remove_background_tiles (bool) – Remove background tiles

Returns:

The shape of the image, the shape of the patches, and the image

Return type:

tuple[tuple[int, int], tuple[int, int], np.ndarray]

liom_toolkit.segmentation.vseg.utils.process_image(image: ndarray, device: device) → Tensor

Process an image to present to U-net model

Parameters:

  • image (np.ndarray) – The image to process

  • device (torch.device) – The device to use

Returns:

The processed image

Return type:

torch.Tensor

liom_toolkit.segmentation.vseg.utils.seeding(seed: int) → None

Set seed for reproducibility

Parameters:

seed (int) – The seed to set

Returns:

None