smoothrandaugment/smooth_geometric_augmentations.py at main · diego-moranda/smoothrandaugment · GitHub

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
import torch.nn.functional as F
import torch
from torch import Tensor
from smooth_augmentation import SmoothAugmentation

class SmoothGeometricAugmentation(SmoothAugmentation):

    def __init__(self, aug_p=1, n_frmaes=150, min_bound=0.0, max_bound=1.0, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Geometric augmentation base class.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum value for the augmentation. Defaults to 0.0.
            max_bound (float, optional): Maximum value for the augmentation. Defaults to 1.0.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave)
        self.multiple_augs = multiple_augs

    def smooth_operation(self, vid) -> Tensor:

        p = torch.rand(1)

        if p > self.aug_p:
            return vid

        vid = vid.unsqueeze(0) if len(vid.shape) < 5 else vid

        factors = self.gen_factor_func()

        aff_mat = self.get_affine_matrix(factors)

        grid = F.affine_grid(aff_mat, vid[0].size(), align_corners=True)
        vid = F.grid_sample(vid[0], grid, padding_mode="zeros", align_corners=True)

        return vid if self.multiple_augs else vid.transpose(1, 0)

    def get_affine_matrix(self, factors: Tensor) -> Tensor:
        """Get the affine matrix used by a geometric augmentation.

        Args:
            factors (Tensor): Generated factors used by the affine matrix

        Returns:
            Tensor: Affine matrix
        """
        pass

class SmoothRotation(SmoothGeometricAugmentation):
    def __init__(self, aug_p=1, n_frmaes=150, min_bound=-30.0, max_bound=30.0, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Smooth Rotation.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum angle in degrees for the augmentation. Defaults to -30.0.
            max_bound (float, optional): Maximum angle in degrees the augmentation. Defaults to 30.0.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave, multiple_augs)

    def get_affine_matrix(self, factors: Tensor) -> Tensor:
        angles = factors / 180 * torch.pi
        s = torch.sin(angles)
        c = torch.cos(angles)

        aff_mat = torch.stack((torch.stack([c, -s, torch.zeros(self.n_frames)], dim=1),
                            torch.stack([s, c, torch.zeros(self.n_frames)], dim=1)), dim=1)
        return aff_mat

class SmoothZoom(SmoothGeometricAugmentation):
    def __init__(self, aug_p=1, n_frmaes=150, min_bound=0.5, max_bound=2.0, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Smooth Zoom.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum zoom factor for the augmentation. Defaults to 0.5.
            max_bound (float, optional): Maximum zoom factor the augmentation. Defaults to 2.0.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave, multiple_augs)

    def get_affine_matrix(self, factors:Tensor) -> Tensor:
        aff_mat = torch.stack((torch.stack([torch.ones(self.n_frames)/factors, torch.zeros(self.n_frames), torch.zeros(self.n_frames)], dim=1),
                            torch.stack([torch.zeros(self.n_frames), torch.ones(self.n_frames)/factors, torch.zeros(self.n_frames)], dim=1)), dim=1)
        return aff_mat

class SmoothTranslateX(SmoothGeometricAugmentation):
    def __init__(self, aug_p=1, n_frmaes=150, min_bound=-1.0, max_bound=1.0, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Smooth Translate on X axis.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum translate factor for the augmentation. Defaults to -1.0.
            max_bound (float, optional): Maximum translate factor the augmentation. Defaults to 1.0.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave, multiple_augs)

    def get_affine_matrix(self, factors:Tensor) -> Tensor:
        aff_mat = torch.stack((torch.stack([torch.ones(self.n_frames), torch.zeros(self.n_frames), factors], dim=1),
                            torch.stack([torch.zeros(self.n_frames), torch.ones(self.n_frames), torch.zeros(self.n_frames)], dim=1)), dim=1)
        return aff_mat

class SmoothTranslateY(SmoothGeometricAugmentation):
    def __init__(self, aug_p=1, n_frmaes=150, min_bound=-1.0, max_bound=1.0, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Smooth Translate on Y axis.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum translate factor for the augmentation. Defaults to -1.0.
            max_bound (float, optional): Maximum translate factor the augmentation. Defaults to 1.0.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave, multiple_augs)

    def get_affine_matrix(self, factors:Tensor) -> Tensor:
        aff_mat = torch.stack((torch.stack([torch.ones(self.n_frames), torch.zeros(self.n_frames), torch.zeros(self.n_frames)], dim=1),
                            torch.stack([torch.zeros(self.n_frames), torch.ones(self.n_frames), factors], dim=1)), dim=1)
        return aff_mat

class SmoothShearX(SmoothGeometricAugmentation):
    def __init__(self, aug_p=1, n_frmaes=150, min_bound=-0.5, max_bound=0.5, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Smooth Shear on X axis.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum shear factor for the augmentation. Defaults to -0.5.
            max_bound (float, optional): Maximum shear factor the augmentation. Defaults to 0.5.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave, multiple_augs)

    def get_affine_matrix(self, factors:Tensor) -> Tensor:
        aff_mat = torch.stack((torch.stack([torch.ones(self.n_frames), factors, torch.zeros(self.n_frames)], dim=1),
                            torch.stack([torch.zeros(self.n_frames), torch.ones(self.n_frames), torch.zeros(self.n_frames)], dim=1)), dim=1)
        return aff_mat

class SmoothShearY(SmoothGeometricAugmentation):
    def __init__(self, aug_p=1, n_frmaes=150, min_bound=-0.5, max_bound=0.5, linear=True, n_signals=50, weighted_wave=False, multiple_augs=False) -> None:
        """Smooth Shear on Y axis.

        Args:
            aug_p (float, optional): Agumentation probability. Defaults to 1.0.
            n_frmaes (int, optional): Number of frames of the videos. Defaults to 150.
            min_bound (float, optional): Minimum shear factor for the augmentation. Defaults to -0.5.
            max_bound (float, optional): Maximum shear factor the augmentation. Defaults to 0.5.
            linear (bool, optional): Whether to use linear factors (True) or generated from sine waves (False). Defaults to True.
            n_signals (int, optional): Number of sinusoidal signals to be used when using non linear technique (linear=False). Defaults to 50.
            weighted_wave (bool, optional): Whether to use a random weight that divides the generated factors of the sinusoidal technique. Defaults to False.
            multiple_augs (bool, optional): Set this to true when chaining multiple geometric augmentations. Don't set to True for the last augmentation. Defaults to False.
        """
        super().__init__(aug_p, n_frmaes, min_bound, max_bound, linear, n_signals, weighted_wave, multiple_augs)

    def get_affine_matrix(self, factors:Tensor) -> Tensor:
        aff_mat = torch.stack((torch.stack([torch.ones(self.n_frames), torch.zeros(self.n_frames), torch.zeros(self.n_frames)], dim=1),
                            torch.stack([factors, torch.ones(self.n_frames), torch.zeros(self.n_frames)], dim=1)), dim=1)
        return aff_mat