File size: 7,364 Bytes
5fc3d65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn.functional as F
from math import sin, cos, pi
import numbers
import random

"""
    Data augmentation functions.

    There are some problems with torchvision data augmentation functions:
    1. they work only on PIL images, which means they cannot be applied to tensors with more than 3 channels,
       and they require a lot of conversion from Numpy -> PIL -> Tensor

    2. they do not provide access to the internal transformations (affine matrices) used, which prevent
       applying them for more complex tasks, such as transformation of an optic flow field (for which
       the inverse transformation must be known).

    For these reasons, we implement my own data augmentation functions
    (strongly inspired by torchvision transforms) that operate directly
    on Torch Tensor variables, and that allow to transform an optic flow field as well.
"""


class Compose(object):
    """Composes several transforms together.
    Args:
        transforms (list of ``Transform`` objects): list of transforms to compose.
    Example:
        >>> transforms.Compose([
        >>>     transforms.CenterCrop(10),
        >>>     transforms.ToTensor(),
        >>> ])
    """

    def __init__(self, transforms):
        self.transforms = transforms

    def __call__(self, x, is_flow=False):
        for t in self.transforms:
            x = t(x, is_flow)
        return x

    def __repr__(self):
        format_string = self.__class__.__name__ + '('
        for t in self.transforms:
            format_string += '\n'
            format_string += '    {0}'.format(t)
        format_string += '\n)'
        return format_string


class CenterCrop(object):
    """Center crop the tensor to a certain size.
    """

    def __init__(self, size, preserve_mosaicing_pattern=False):
        if isinstance(size, numbers.Number):
            self.size = (int(size), int(size))
        else:
            self.size = size

        self.preserve_mosaicing_pattern = preserve_mosaicing_pattern

    def __call__(self, x, is_flow=False):
        """
            x: [C x H x W] Tensor to be rotated.
            is_flow: this parameter does not have any effect
        Returns:
            Tensor: Cropped tensor.
        """
        
        w, h = x.shape[2], x.shape[1]
        th, tw = self.size
        # print(tw, w)
        # print(th, h)
        assert(th <= h)
        assert(tw <= w)
        i = int(round((h - th) / 2.))
        j = int(round((w - tw) / 2.))

        if self.preserve_mosaicing_pattern:
            # make sure that i and j are even, to preserve
            # the mosaicing pattern
            if i % 2 == 1:
                i = i + 1
            if j % 2 == 1:
                j = j + 1

        return x[:, i:i + th, j:j + tw]

    def __repr__(self):
        return self.__class__.__name__ + '(size={0})'.format(self.size)


class RandomCrop(object):
    """Crop the tensor at a random location.
    """

    def __init__(self, size, preserve_mosaicing_pattern=False):
        if isinstance(size, numbers.Number):
            self.size = (int(size), int(size))
        else:
            self.size = size

        self.preserve_mosaicing_pattern = preserve_mosaicing_pattern

    @staticmethod
    def get_params(x, output_size):
        w, h = x.shape[2], x.shape[1]
        th, tw = output_size
        assert(th <= h)
        assert(tw <= w)
        if w == tw and h == th:
            return 0, 0, h, w

        i = random.randint(0, h - th)
        j = random.randint(0, w - tw)

        return i, j, th, tw

    def __call__(self, x, is_flow=False):
        """
            x: [C x H x W] Tensor to be rotated.
            is_flow: this parameter does not have any effect
        Returns:
            Tensor: Cropped tensor.
        """
        i, j, h, w = self.get_params(x, self.size)

        if self.preserve_mosaicing_pattern:
            # make sure that i and j are even, to preserve the mosaicing pattern
            if i % 2 == 1:
                i = i + 1
            if j % 2 == 1:
                j = j + 1

        return x[:, i:i + h, j:j + w]

    def __repr__(self):
        return self.__class__.__name__ + '(size={0})'.format(self.size)


class RandomRotationFlip(object):
    """Rotate the image by angle.
    """

    def __init__(self, degrees, p_hflip=0.5, p_vflip=0.5):
        if isinstance(degrees, numbers.Number):
            if degrees < 0:
                raise ValueError("If degrees is a single number, it must be positive.")
            self.degrees = (-degrees, degrees)
        else:
            if len(degrees) != 2:
                raise ValueError("If degrees is a sequence, it must be of len 2.")
            self.degrees = degrees

        self.p_hflip = p_hflip
        self.p_vflip = p_vflip

    @staticmethod
    def get_params(degrees, p_hflip, p_vflip):
        """Get parameters for ``rotate`` for a random rotation.
        Returns:
            sequence: params to be passed to ``rotate`` for random rotation.
        """
        angle = random.uniform(degrees[0], degrees[1])
        angle_rad = angle * pi / 180.0

        M_original_transformed = torch.FloatTensor([[cos(angle_rad), -sin(angle_rad), 0],
                                                    [sin(angle_rad), cos(angle_rad), 0],
                                                    [0, 0, 1]])

        if random.random() < p_hflip:
            M_original_transformed[:, 0] *= -1

        if random.random() < p_vflip:
            M_original_transformed[:, 1] *= -1

        M_transformed_original = torch.inverse(M_original_transformed)

        M_original_transformed = M_original_transformed[:2, :].unsqueeze(dim=0)  # 3 x 3 -> N x 2 x 3
        M_transformed_original = M_transformed_original[:2, :].unsqueeze(dim=0)

        return M_original_transformed, M_transformed_original

    def __call__(self, x, is_flow=False):
        """
            x: [C x H x W] Tensor to be rotated.
            is_flow: if True, x is an [2 x H x W] displacement field, which will also be transformed
        Returns:
            Tensor: Rotated tensor.
        """
        assert(len(x.shape) == 3)

        if is_flow:
            assert(x.shape[0] == 2)

        M_original_transformed, M_transformed_original = self.get_params(self.degrees, self.p_hflip, self.p_vflip)
        affine_grid = F.affine_grid(M_original_transformed, x.unsqueeze(dim=0).shape, align_corners=False)
        transformed = F.grid_sample(x.unsqueeze(dim=0), affine_grid, align_corners=False)

        if is_flow:
            # Apply the same transformation to the flow field
            A00 = M_transformed_original[0, 0, 0]
            A01 = M_transformed_original[0, 0, 1]
            A10 = M_transformed_original[0, 1, 0]
            A11 = M_transformed_original[0, 1, 1]
            vx = transformed[:, 0, :, :].clone()
            vy = transformed[:, 1, :, :].clone()
            transformed[:, 0, :, :] = A00 * vx + A01 * vy
            transformed[:, 1, :, :] = A10 * vx + A11 * vy

        return transformed.squeeze(dim=0)

    def __repr__(self):
        format_string = self.__class__.__name__ + '(degrees={0}'.format(self.degrees)
        format_string += ', p_flip={:.2f}'.format(self.p_hflip)
        format_string += ', p_vlip={:.2f}'.format(self.p_vflip)
        format_string += ')'
        return format_string