Added a new module for future optional data augmentation methods

pytorch/transforms.py

+# -*- coding: utf-8 -*-
+"""
+Created on Fri Jul 17 15:28:18 2020
+
+@author: Daniel
+"""
+
+# externals
+import numpy as np
+from scipy import ndimage
+
+
+class Transformation(object):
+
+    def __init__(self):
+        pass
+
+    def apply(self, image):
+        raise NotImplementedError
+
+
+class VariantTransformation(Transformation):
+
+    def __init__(self):
+
+        # requires transformation on the ground thruth
+        self.invariant = False
+
+
+class InvariantTransformation(Transformation):
+
+    def __init__(self):
+
+        # transformation on the ground truth not required
+        self.invariant = True
+
+
+class FlipLr(VariantTransformation):
+
+    def __init__(self):
+        super().__init__()
+
+    def apply(self, image):
+        return np.asarray(image)[..., ::-1]
+
+
+class FlipUd(VariantTransformation):
+
+    def __init__(self):
+        super().__init__()
+
+    def _transform(self, image):
+        return np.asarray(image)[..., ::-1, :]
+
+
+class Rotate(VariantTransformation):
+
+    def __init__(self):
+        super().__init__()
+
+    def apply(self, image, angle):
+
+        # check dimension of input image
+        ndim = np.asarray(image).ndim
+
+        # axes defining the rotational plane
+        rot_axes = (0, 1)
+        if ndim > 2:
+            rot_axes = (ndim - 2, ndim - 1)
+
+        return ndimage.rotate(image, angle, axes=rot_axes, reshape=False)