Datasets

datasets/__init__.py

@@ -0,0 +1,7 @@
+from .dataset import Dataset, Dataloader, Cephalogram
+
+from .aariz import AarizDataset
+from .isbi import ISBIDataset
+from .pku import PKUDataset
+
+from .paths import Paths

datasets/paths.py

@@ -0,0 +1,16 @@
+import os
+
+class Paths(object):
+    root = "path\to\datasets\folder"
+
+    @classmethod
+    def dataset_root_path(cls, name: str=None):
+
+        if name == "isbi":
+            return os.path.join(cls.root, "ISBI")
+        elif name == "pku":
+            return os.path.join(cls.root, "PKU")
+        elif name == "aariz":
+            return os.path.join(cls.root, "Aariz")
+        else:
+            raise ValueError("\'{}\' dataset doesn't exists in your paths.py file".format(name))

datasets/utils.py

@@ -0,0 +1,82 @@
+import numpy as np
+import cv2
+
+def calculate_new_dimensions(height, width, max_size: int=512):
+    aspect_ratio = width / height
+    if height > width:
+        height = max_size
+        width  = int(height * aspect_ratio)
+    else:
+        width  = max_size
+        height = int(width / aspect_ratio)
+
+    return height, width
+
+def resize(
+    image: np.ndarray,
+    landmarks: np.ndarray=None,
+    dimensions: tuple=(int, int)
+):
+    image_height, image_width = image.shape[0:2]
+    new_height, new_width = dimensions
+    ratio_height, ratio_width = (image_height / new_height), (image_width / new_width)
+
+    image = cv2.resize(
+        image,
+        (new_width, new_height),
+        interpolation=cv2.INTER_AREA
+    )
+
+    if landmarks is not None:
+        landmarks = np.stack([
+            landmarks[:, 0] / ratio_width,
+            landmarks[:, 1] / ratio_height
+        ], axis=-1)
+
+        return image, landmarks
+
+    return image
+
+def normalize_landmarks(
+    landmarks: np.ndarray, 
+    height: int, 
+    width: int, 
+    num_landmarks: int=19
+):
+    original_shape = landmarks.shape
+    landmarks = np.reshape(landmarks, (-1, num_landmarks, 2))
+
+    landmarks = np.stack([
+        landmarks[:, :, 0] / width,
+        landmarks[:, :, 1] / height
+    ], axis=-1, dtype=np.float32)
+
+    if len(original_shape) == 2:
+        landmarks = np.reshape(landmarks, original_shape)
+
+    return landmarks
+
+def denormalize_landmarks(
+    landmarks: np.ndarray, 
+    height: int, 
+    width: int, 
+    num_landmarks: int=19
+):
+    original_shape = landmarks.shape
+    landmarks = np.reshape(landmarks, (-1, num_landmarks, 2))
+
+    landmarks = np.stack([
+        landmarks[:, :, 0] * width,
+        landmarks[:, :, 1] * height
+    ], axis=-1, dtype=np.float32)
+
+    if len(original_shape) == 2:
+        landmarks = np.reshape(landmarks, original_shape)
+
+    return landmarks.round()
+
+def rescale(image, scale: float, offset: int = 0, dtype: str="float32"):
+    scale = np.array(scale, dtype)
+    offset = np.array(offset, dtype)
+    image = image * scale + offset
+    return image