File size: 8,301 Bytes
b5cbaa6
8e6512c
 
 
b5cbaa6
 
8e6512c
b5cbaa6
 
8e6512c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b5cbaa6
 
 
 
 
8e6512c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b5cbaa6
 
 
 
8e6512c
b5cbaa6
8e6512c
 
 
 
 
 
 
 
 
 
b5cbaa6
 
 
 
 
 
8e6512c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b5cbaa6
 
 
 
8e6512c
b5cbaa6
8e6512c
 
 
 
 
 
 
 
 
b5cbaa6
 
 
 
 
8e6512c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b5cbaa6
 
8e6512c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
import random

import numpy as np
from PIL import Image, ImageEnhance, ImageOps
from torchvision import transforms


class TNetPolicy(object):
    """
    Applies the augmentation policy used in Jun et al's coronary artery segmentation T-Net
    https://arxiv.org/abs/1905.04197. As described by the authors, first they zoom-in or zoom-out at a
    random ratio within +/- 20%. Then, the image is shifted, horizontally and vertically, at a random
    ratio within +/- 20% of the image size (512 x 512). Then, the angiography is rotated between
    +/- 30 degrees. The rotation angle is not larger because actual angiographies do not deviate much
    from this range. Finally, because the brightness of the angiography image can vary, the brightness is
    also changed within +/- 40% at random rates.
    Referred in the results as Aug1.
    Example:
    >>> policy = TNetPolicy()
    >>> transformed_img, transformed_mask = policy(image, mask)
    """

    def __init__(
        self,
        scale_ranges=[0.8, 1.2],
        img_size=[512, 512],
        translate=[0.2, 0.2],
        rotation=[-30, 30],
        brightness=0.4,
    ):
        self.scale_ranges = scale_ranges
        self.img_size = img_size
        self.translate = translate
        self.rotation = rotation
        self.brightness = brightness

    def __call__(self, image, mask=None):
        tf_mask = None
        tf_list = list()  # List of transformation

        # Random zoom-in or zoom-out of -20% to 20%
        params = transforms.RandomAffine.get_params(
            degrees=[0, 0],
            translate=[0, 0],
            scale_ranges=self.scale_ranges,
            img_size=self.img_size,
            shears=[0, 0],
        )
        tf_image = transforms.functional.affine(
            image, params[0], params[1], params[2], params[3]
        )
        if mask is not None:
            tf_mask = transforms.functional.affine(
                mask, params[0], params[1], params[2], params[3]
            )

        # Random horizontal and vertical shift of -20% to 20%
        params = transforms.RandomAffine.get_params(
            degrees=[0, 0],
            translate=self.translate,
            scale_ranges=[1, 1],
            img_size=self.img_size,
            shears=[0, 0],
        )
        tf_image = transforms.functional.affine(
            tf_image, params[0], params[1], params[2], params[3]
        )
        if mask is not None:
            tf_mask = transforms.functional.affine(
                tf_mask, params[0], params[1], params[2], params[3]
            )

        # Random rotation of -30 to 30 degress
        angle = transforms.RandomRotation.get_params(self.rotation)
        tf_image = transforms.functional.rotate(tf_image, angle)
        if mask is not None:
            tf_mask = transforms.functional.rotate(tf_mask, angle)

        # Random brightness change of -40% to 40%
        tf = transforms.ColorJitter(brightness=self.brightness)
        tf_image = tf(tf_image)

        if mask is not None:
            return (tf_image, tf_mask)
        else:
            return tf_image

    def __repr__(self):
        return "TNet Coronary Artery Segmentation Augmentation Policy"


class RetinaPolicy(object):
    def __init__(
        self,
        scale_ranges=[1, 1.1],
        img_size=[512, 512],
        translate=[0.1, 0.1],
        rotation=[-20, 20],
        crop_dims=[480, 480],
        brightness=None,
    ):
        self.scale_ranges = scale_ranges
        self.img_size = img_size
        self.translate = translate
        self.rotation = rotation
        self.brightness = brightness
        self.crop_dims = crop_dims

    def __call__(self, image, mask=None):
        tf_mask = None

        # Random crop
        i, j, h, w = transforms.RandomCrop.get_params(image, self.crop_dims)
        tf_image = transforms.functional.crop(image, i, j, h, w)
        if mask is not None:
            tf_mask = transforms.functional.crop(mask, i, j, h, w)

        # Random rotation of -20 to 20 degress
        angle = transforms.RandomRotation.get_params(self.rotation)
        tf_image = transforms.functional.rotate(tf_image, angle)
        if mask is not None:
            tf_mask = transforms.functional.rotate(tf_mask, angle)

        # Random horizontal and vertical shift of -10% to 10%
        params = transforms.RandomAffine.get_params(
            degrees=[0, 0],
            translate=self.translate,
            scale_ranges=[1, 1],
            img_size=self.img_size,
            shears=[0, 0],
        )
        tf_image = transforms.functional.affine(
            tf_image, params[0], params[1], params[2], params[3]
        )
        if mask is not None:
            tf_mask = transforms.functional.affine(
                tf_mask, params[0], params[1], params[2], params[3]
            )

        # TODO: -10% to 10% may make more sense, due to the existance of images with black padding borders
        # Random zoom-in of 0% to 10%
        params = transforms.RandomAffine.get_params(
            degrees=[0, 0],
            translate=[0, 0],
            scale_ranges=self.scale_ranges,
            img_size=self.img_size,
            shears=[0, 0],
        )
        tf_image = transforms.functional.affine(
            tf_image, params[0], params[1], params[2], params[3]
        )
        if mask is not None:
            tf_mask = transforms.functional.affine(
                tf_mask, params[0], params[1], params[2], params[3]
            )

        # TODO: change brightness too
        # Random brightness change
        if self.brightness is not None:
            tf = transforms.ColorJitter(brightness=self.brightness)
            tf_image = tf(tf_image)

        if mask is not None:
            return (tf_image, tf_mask)
        else:
            return tf_image

    def __repr__(self):
        return "Retinal Vessel Segmentation Augmentation Policy"


class CoronaryPolicy(object):
    def __init__(
        self,
        scale_ranges=[1, 1.1],
        img_size=[512, 512],
        translate=[0.1, 0.1],
        rotation=[-20, 20],
        brightness=None,
    ):
        self.scale_ranges = scale_ranges
        self.img_size = img_size
        self.translate = translate
        self.rotation = rotation
        self.brightness = brightness

    def __call__(self, image, mask=None):
        tf_mask = None
        # Random rotation of -20 to 20 degress
        angle = transforms.RandomRotation.get_params(self.rotation)
        tf_image = transforms.functional.rotate(image, angle)
        if mask is not None:
            tf_mask = transforms.functional.rotate(mask, angle)

        # Random horizontal and vertical shift of -10% to 10%
        params = transforms.RandomAffine.get_params(
            degrees=[0, 0],
            translate=self.translate,
            scale_ranges=[1, 1],
            img_size=self.img_size,
            shears=[0, 0],
        )
        tf_image = transforms.functional.affine(
            tf_image, params[0], params[1], params[2], params[3]
        )
        if mask is not None:
            tf_mask = transforms.functional.affine(
                tf_mask, params[0], params[1], params[2], params[3]
            )

        # TODO: -10% to 10% may make more sense, due to the existance of images with black padding borders
        # Random zoom-in of 0% to 10%
        params = transforms.RandomAffine.get_params(
            degrees=[0, 0],
            translate=[0, 0],
            scale_ranges=self.scale_ranges,
            img_size=self.img_size,
            shears=[0, 0],
        )
        tf_image = transforms.functional.affine(
            tf_image, params[0], params[1], params[2], params[3]
        )
        if mask is not None:
            tf_mask = transforms.functional.affine(
                tf_mask, params[0], params[1], params[2], params[3]
            )

        # TODO: change brightness too
        # Random brightness change
        if self.brightness is not None:
            tf = transforms.ColorJitter(brightness=self.brightness)
            tf_image = tf(tf_image)

        if mask is not None:
            return (tf_image, tf_mask)
        else:
            return tf_image

    def __repr__(self):
        return "Coronary Artery Segmentation Augmentation Policy"