Utils Module#

Utils module.

landmarker.utils.all_annotations_to_landmarks(paths, class_names)[source]#

Convert all annotations of LabelMe to landmarks.

Parameters:

  • paths (list[str]) – list of paths to the annotations

  • class_names (list[str]) – class names

Returns:

landmarks

Return type:

torch.Tensor

landmarker.utils.all_annotations_to_landmarks_numpy(paths, class_names)[source]#

Convert all annotations of LabelMe to landmarks (Numpy version).

Parameters:

  • paths (list[str]) – list of paths to the annotations

  • class_names (list[str]) – class names

Returns:

landmarks

Return type:

np.ndarray

landmarker.utils.annotation_to_landmark(json_obj, class_names)[source]#

Convert the annotation of LabelMe to landmarks.

Parameters:

  • json_obj (dict) – annotation in json format

  • class_names (list[str]) – class names

Returns:

landmarks

Return type:

torch.Tensor

landmarker.utils.annotation_to_landmark_numpy(json_obj, class_names)[source]#

Convert the annotation of LabelMe to landmarks (Numpy version).

Parameters:

  • json_obj (dict) – annotation in json format

  • class_names (list[str]) – class names

Returns:

landmarks

Return type:

np.ndarray

landmarker.utils.covert_video_to_frames(video_path, frames_path, zero_fill=6)[source]#

Convert a video to frames.

Parameters:

  • video_path (str) – path to the video

  • frames_path (str) – path to the folder where the frames are saved

  • zero_fill (int, optional) – number of digits in the frame name. Defaults to 6.

Returns:

None

Return type:

None

landmarker.utils.extract_roi(imgs, roi_middle, landmarks, size=256, spatial_dims=2, ensure_dim=True)[source]#

Extract a region of interest around the landmarks in the image.

Parameters:

  • img (torch.Tensor) – Image to extract the region of interest from.

  • roi_middle (torch.Tensor) – Middle of the region of interest.

  • landmark (torch.Tensor) – Landmark to transform to the region of interest.

  • size (int | tuple[int, int]) – Size of the region of interest.

  • spatial_dims (int) – Number of spatial dimensions of the image.

  • ensure_dim (bool) – Ensure the output has the same number of dimensions as the input.

  • imgs (Tensor) –

  • landmarks (Tensor) –

Returns:

Region of interest around the landmarks. torch.Tensor: Landmarks in the region of interest. torch.Tensor: Upper left corner of the region of interest.

Return type:

torch.Tensor

landmarker.utils.get_angle(p1, p2, p3, radial=True)[source]#

Calculate the angle between three points. The angle is calculated in degrees or radians. The angle is calculated in the direction of p2 -> p3. If the angle is radial, the angle is between 0 and 2pi. If the angle is not radial, the angle is between 0 and 360.

Parameters:

  • p1 (torch.Tensor) – first point

  • p2 (torch.Tensor) – second point

  • p3 (torch.Tensor) – third point

  • radial (bool, optional) – whether the angle is radial. Defaults to True.

Returns:

angle

Return type:

torch.Tensor

landmarker.utils.get_angle_numpy(p1, p2, p3, radial=True)[source]#

Calculate the angle between three points (Numpy version). See documentation of get_angle.

Parameters:

  • p1 (np.ndarray) – first point

  • p2 (np.ndarray) – second point

  • p3 (np.ndarray) – third point

  • radial (bool, optional) – whether the angle is radial. Defaults to True.

Returns:

angle

Return type:

np.ndarray

landmarker.utils.get_paths(folder_path, extension)[source]#

Retrieve all paths with a defined extension files recursively from the folder path.

Parameters:

  • folder_path (str) – path to the folder

  • extension (str) – file extension

Returns:

list of paths

Return type:

list[str]

landmarker.utils.normalize(img)[source]#

Normalize image to 0-1 range

Parameters:

img (np.ndarray) – Image to normalize

Returns:

Normalized image

Return type:

np.ndarray

landmarker.utils.pixel_to_unit(landmarks, pixel_spacing=None, dim=None, dim_orig=None, padding=None)[source]#

Convert the landmarks from pixel to unit.

Parameters:

  • landmarks (torch.Tensor) – landmarks

  • pixel_spacing (Optional[torch.Tensor], optional) – pixel spacing. Defaults to None.

  • dim (Optional[tuple[int, ...] | torch.Tensor], optional) – image size. Defaults to None.

  • dim_orig (Optional[torch.Tensor], optional) – original image size. Defaults to None.

  • padding (Optional[torch.Tensor], optional) – padding. Defaults to None.

Returns:

landmarks in units

Return type:

torch.Tensor

landmarker.utils.pixel_to_unit_numpy(landmarks, pixel_spacing=None, dim=None, dim_orig=None, padding=None)[source]#

Convert the landmarks from pixel to unit (Numpy version).

Parameters:

  • landmarks (np.ndarray) – landmarks

  • pixel_spacing (Optional[np.ndarray], optional) – pixel spacing. Defaults to None.

  • dim (Optional[tuple[int, ...] | np.ndarray], optional) – image size. Defaults to None.

  • dim_orig (Optional[np.ndarray], optional) – original image size. Defaults to None.

  • padding (Optional[np.ndarray], optional) – padding. Defaults to None.

Returns:

landmarks in units

Return type:

np.ndarray

landmarker.utils.preprocess_all(folder, output_folder, fun_name)[source]#

Apply a preprocessing function to all images in a folder.

Parameters:

  • folder (str) – Folder containing images to preprocess

  • output_folder (str) – Folder to store the preprocessed images

  • fun_name (str) – Name of the function to apply. (Only normalize supported for now)

Return type:

None