File size: 9,038 Bytes
94f372a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
import torch
from torch import nn


class MLP(nn.Module):
    def __init__(

        self,

        initial_dim=512,

        hidden_dim=[128, 32, 2],

        final_dim=2,

        norm=nn.InstanceNorm1d,

        activation=nn.ReLU,

        aux_data=[],

    ):
        """

        Initializes an MLP Classification Head

        Args:

            hidden_dim (list): list of hidden dimensions for the MLP

            norm (nn.Module): normalization layer

            activation (nn.Module): activation layer

        """
        super().__init__()
        self.aux_data = aux_data
        self.aux = len(self.aux_data) > 0
        if self.aux:
            hidden_dim_aux = hidden_dim
            hidden_dim_aux[-1] = 128
            final_dim_aux_dict = {
                "land_cover": 12,
                "climate": 30,
                "soil": 14,
                "road_index": 1,
                "drive_side": 1,
                "dist_sea": 1,
            }
            self.idx = {}
            final_dim_aux = 0
            for col in self.aux_data:
                self.idx[col] = [
                    final_dim_aux + i for i in range(final_dim_aux_dict[col])
                ]
                final_dim_aux += final_dim_aux_dict[col]
            dim = [initial_dim] + hidden_dim_aux + [final_dim_aux]
            args = self.init_layers(dim, norm, activation)
            self.mlp_aux = nn.Sequential(*args)
        dim = [initial_dim] + hidden_dim + [final_dim]
        args = self.init_layers(dim, norm, activation)
        self.mlp = nn.Sequential(*args)

    def init_layers(self, dim, norm, activation):
        """Initializes the MLP layers."""
        args = [nn.LayerNorm(dim[0])]
        for i in range(len(dim) - 1):
            args.append(nn.Linear(dim[i], dim[i + 1]))
            if i < len(dim) - 2:
                # args.append(norm(dim[i + 1]))
                args.append(norm(4, dim[i + 1]))
                args.append(activation())
        return args

    def forward(self, x):
        """Predicts GPS coordinates from an image.

        Args:

            x: torch.Tensor with features

        """
        if self.aux:
            out = {"gps": self.mlp(x[:, 0, :])}
            x = self.mlp_aux(x[:, 0, :])
            for col in list(self.idx.keys()):
                out[col] = x[:, self.idx[col]]
            return out
        return self.mlp(x[:, 0, :])

class MLPResNet(nn.Module):
    def __init__(

        self,

        initial_dim=512,

        hidden_dim=[128, 32, 2],

        final_dim=2,

        norm=nn.InstanceNorm1d,

        activation=nn.ReLU,

        aux_data=[],

    ):
        """

        Initializes an MLP Classification Head

        Args:

            hidden_dim (list): list of hidden dimensions for the MLP

            norm (nn.Module): normalization layer

            activation (nn.Module): activation layer

        """
        super().__init__()
        self.aux_data = aux_data
        self.aux = len(self.aux_data) > 0
        if self.aux:
            hidden_dim_aux = hidden_dim
            hidden_dim_aux[-1] = 128
            final_dim_aux_dict = {
                "land_cover": 12,
                "climate": 30,
                "soil": 14,
                "road_index": 1,
                "drive_side": 1,
                "dist_sea": 1,
            }
            self.idx = {}
            final_dim_aux = 0
            for col in self.aux_data:
                self.idx[col] = [
                    final_dim_aux + i for i in range(final_dim_aux_dict[col])
                ]
                final_dim_aux += final_dim_aux_dict[col]
            dim = [initial_dim] + hidden_dim_aux + [final_dim_aux]
            args = self.init_layers(dim, norm, activation)
            self.mlp_aux = nn.Sequential(*args)
        dim = [initial_dim] + hidden_dim + [final_dim]
        args = self.init_layers(dim, norm, activation)
        self.mlp = nn.Sequential(*args)

    def init_layers(self, dim, norm, activation):
        """Initializes the MLP layers."""
        args = [nn.LayerNorm(dim[0])]
        for i in range(len(dim) - 1):
            args.append(nn.Linear(dim[i], dim[i + 1]))
            if i < len(dim) - 2:
                # args.append(norm(dim[i + 1]))
                args.append(norm(4, dim[i + 1]))
                args.append(activation())
        return args

    def forward(self, x):
        """Predicts GPS coordinates from an image.

        Args:

            x: torch.Tensor with features

        """
        if self.aux:
            out = {"gps": self.mlp(x[:, 0, :])}
            x = self.mlp_aux(x[:, 0, :])
            for col in list(self.idx.keys()):
                out[col] = x[:, self.idx[col]]
            return out
        return self.mlp(x)


class MLPCentroid(nn.Module):
    def __init__(

        self,

        initial_dim=512,

        hidden_dim=[128, 32, 2],

        final_dim=2,

        norm=nn.InstanceNorm1d,

        activation=nn.ReLU,

        aux_data=[],

    ):
        """

        Initializes an MLP Classification Head

        Args:

            hidden_dim (list): list of hidden dimensions for the MLP

            norm (nn.Module): normalization layer

            activation (nn.Module): activation layer

        """
        super().__init__()
        self.aux_data = aux_data
        self.aux = len(self.aux_data) > 0
        dim = [initial_dim] + hidden_dim + [final_dim // 3]
        args = self.init_layers(dim, norm, activation)
        self.classif = nn.Sequential(*args)
        dim = [initial_dim] + hidden_dim + [2 * final_dim // 3]
        args = self.init_layers(dim, norm, activation)
        self.reg = nn.Sequential(*args)
        # torch.nn.init.normal_(self.reg.weight, mean=0.0, std=0.01)
        if self.aux:
            self.dim = [initial_dim] + hidden_dim
            self.predictors = {"gps": self.mlp}
            self.init_aux(dim, norm, activation)

    def init_layers(self, dim, norm, activation):
        """Initializes the MLP layers."""
        args = [nn.LayerNorm(dim[0])]
        for i in range(len(dim) - 1):
            args.append(nn.Linear(dim[i], dim[i + 1]))
            if i < len(dim) - 2:
                # args.append(norm(dim[i + 1]))
                args.append(norm(4, dim[i + 1]))
                args.append(activation())
        return args

    def init_aux(self, dim, norm, activation):
        final_dim_aux = {
            "land_cover": 12,
            "climate": 30,
            "soil": 14,
            "road_index": 1,
            "drive_side": 1,
            "dist_sea": 1,
        }
        if "land_cover" in self.aux_data:
            args = self.init_layers(
                self.dim + [final_dim_aux["land_cover"]], norm, activation
            )
            self.land_cover = nn.Sequential(*args)
            self.predictors["land_cover"] = self.land_cover
        if "road_index" in self.aux_data:
            args = self.init_layers(
                self.dim + [final_dim_aux["road_index"]], norm, activation
            )
            self.road_index = nn.Sequential(*args)
            self.predictors["road_index"] = self.road_index
        if "drive_side" in self.aux_data:
            args = self.init_layers(
                self.dim + [final_dim_aux["drive_side"]], norm, activation
            )
            self.drive_side = nn.Sequential(*args)
            self.predictors["drive_side"] = self.drive_side
        if "climate" in self.aux_data:
            args = self.init_layers(
                self.dim + [final_dim_aux["climate"]], norm, activation
            )
            self.climate = nn.Sequential(*args)
            self.predictors["climate"] = self.climate
        if "soil" in self.aux_data:
            args = self.init_layers(
                self.dim + [final_dim_aux["soil"]], norm, activation
            )
            self.soil = nn.Sequential(*args)
            self.predictors["soil"] = self.soil
        if "dist_sea" in self.aux_data:
            args = self.init_layers(
                self.dim + [final_dim_aux["dist_sea"]], norm, activation
            )
            self.dist_sea = nn.Sequential(*args)
            self.predictors["dist_sea"] = self.dist_sea

    def forward(self, x):
        """Predicts GPS coordinates from an image.

        Args:

            x: torch.Tensor with features

        """
        if self.aux:
            return {
                col: self.predictors[col](x[:, 0, :]) for col in self.predictors.keys()
            }
        return torch.cat([self.classif(x[:, 0, :]), self.reg(x[:, 0, :])], dim=1)


class Identity(nn.Module):
    def __init__(

        self

    ):
        """

        Initializes an Identity module

        """
        super().__init__()

    def forward(self, x):
        """

        Return same as input

        """
        return x