SpacEX/ttorch_datamodule.py at main · ahmedemam576/SpacEX · 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
import os
import random
import warnings

import torch
import torch.nn.functional as F
import numpy as np
import pandas as pd

from tlib import ttorch


class AnthroProtectDataset(ttorch.data.images.Dataset):

    def __init__(self, **kwargs):
        super().__init__(**kwargs)


class AnthroProtectDataModule(ttorch.data.images.DataModule):

    DatasetClass = AnthroProtectDataset

    def __init__(self, csv_file, folder, channels, batch_size, num_workers):

        file_infos_df = pd.read_csv(csv_file)
        #file_infos_df = file_infos_df[file_infos_df['label'] == 1]  # only protected areas

        super().__init__(
            file_infos_df=file_infos_df,
            folder=folder,

            channels=channels,
            x_normalization=(0, 10000),
            clip_range=(0, 1),
            rotate=False,
            cutmix=None,
            n_classes=1,

            use_rasterio=True,
            rgb_channels=[2, 1, 0],

            batch_size=batch_size,
            num_workers=num_workers,
        )


class MapInWildDataset(ttorch.data.images.Dataset):

    def transform(self, x, y):
        """
        Transforms given x and y.

        :param x: input data
        :param y: target data
        :return: tuple of transformed data (x, y)
        """

        # take only defined channels
        if self.channels is not None:
            x = x[self.channels]

        # cutmix
        if self.cutmix is not None and random.uniform(0, 1) <= self.cutmix and self.training:  # perform with given probability

            #indices = range(len(self.files))
            indices = [i for i in range(len(self.files)) if self.labels[i] != y]

            # get second item
            index2 = random.choice(indices)
            batch2 = self.tgetitem(index2)
            x2, y2 = ttorch.utils.get_batch(batch2)
            if self.channels is not None:
                x2 = x2[self.channels]

            # combine items
            edge = random.choice(['top', 'bottom', 'left', 'right'])
            amount = random.uniform(0, 0.5)

            x = ttorch.data.images.perform_cutmix(t1=x, t2=x2, edge=edge, amount=amount, clone=False)
            y = y * (1 - amount) + y2 * amount

        # x to torch
        x = x.astype(np.single)
        x = torch.tensor(x)

        # x normalization
        if self.x_normalization is not None:
            x = (x - self.x_normalization[0]) / self.x_normalization[1]

        if self.clip_range is not None:
            x =  torch.clip(x, min=self.clip_range[0], max=self.clip_range[1])

        # resize
        if self.resize is not None:
            x = self.resize(x)

        # crop
        if self.crop is not None:
            x = self.crop(x)
            if self.cutmix is not None:
                print(self.cutmix)
                warnings.warn('WARNING: You have defined CutMix as well as cropping. Note that CutMix is performed first.')

        # x rotation
        if self.rotate and self.training:
            x = ttorch.data.images.rotate_randomly_90(x)

        # y to torch
        y = torch.tensor(y)

        # y normalization
        if self.y_normalization is not None:
            y = (y - self.y_normalization[0]) / self.y_normalization[1]

        # y adapt to number of classes
        if self.n_classes is not None:

            if self.n_classes > 1:
                y = int(y)
                # one hot encoding only if multiple classes, else unsqueeze to shape[batch_size] --> shape[batch_size, 1]
                y = F.one_hot(y, self.n_classes).float()
            else:
                y = torch.unsqueeze(y, -1).float()

        return x, y


class MapInWildDataModule(ttorch.data.images.DataModule):

    DatasetClass = MapInWildDataset

    def __init__(self, csv_file, folder, channels, batch_size, num_workers):

        file_infos_df = pd.read_csv(csv_file)
        #file_infos_df = file_infos_df[file_infos_df['label'] == 1]  # only protected areas
        file_infos_df = file_infos_df[file_infos_df['subset'] == True]
        #file_infos_df = file_infos_df[file_infos_df['season'] == 'summer']

        super().__init__(
            file_infos_df=file_infos_df,
            folder=folder,

            channels=channels,
            x_normalization=(0, 10000),
            clip_range=(0, 1),
            rotate=False,
            cutmix=1,
            n_classes=1,

            use_rasterio=True,
            rgb_channels=[2, 1, 0],

            batch_size=batch_size,
            num_workers=num_workers,
        )