File size: 8,622 Bytes
04c78c7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Adapted from monodepth2
# https://github.com/nianticlabs/monodepth2/blob/master/networks/depth_decoder.py
#
# Copyright Niantic 2019. Patent Pending. All rights reserved.
#
# This software is licensed under the terms of the Monodepth2 licence
# which allows for non-commercial use only, the full terms of which are made
# available in the LICENSE file.

from __future__ import absolute_import, division, print_function
from collections import OrderedDict
from easydict import EasyDict

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision.models as models
import torch.utils.model_zoo as model_zoo


class ConvBlock(torch.nn.Module):
    """Layer to perform a convolution followed by ELU."""
    def __init__(self, in_channels, out_channels, bn=False, dropout=0.0):
        super(ConvBlock, self).__init__()

        self.block = nn.Sequential(
            Conv3x3(in_channels, out_channels),
            nn.BatchNorm2d(out_channels) if bn else nn.Identity(),
            nn.ELU(inplace=True),
            # Pay attention: 2d version of dropout is used
            nn.Dropout2d(dropout) if dropout > 0 else nn.Identity())

    def forward(self, x):
        out = self.block(x)
        return out


class Conv3x3(nn.Module):
    """Layer to pad and convolve input with 3x3 kernels."""
    def __init__(self, in_channels, out_channels, use_refl=True):
        super(Conv3x3, self).__init__()

        if use_refl:
            self.pad = nn.ReflectionPad2d(1)
        else:
            self.pad = nn.ZeroPad2d(1)
        self.conv = nn.Conv2d(int(in_channels), int(out_channels), 3)

    def forward(self, x):
        out = self.pad(x)
        out = self.conv(out)
        return out

def upsample(x):
    """Upsample input tensor by a factor of 2."""
    return F.interpolate(x, scale_factor=2, mode="nearest")


class ResNetMultiImageInput(models.ResNet):
    """Constructs a resnet model with varying number of input images.
    Adapted from https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py
    """
    def __init__(self, block, layers, num_classes=1000, in_channels=3):
        super(ResNetMultiImageInput, self).__init__(block, layers)
        self.inplanes = 64
        self.conv1 = nn.Conv2d(
            in_channels, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)


def resnet_multiimage_input(num_layers, pretrained=False, in_channels=3):
    """Constructs a ResNet model.
    Args:
        num_layers (int): Number of resnet layers. Must be 18 or 50
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        in_channels (int): Number of input channels
    """
    assert num_layers in [18, 50], "Can only run with 18 or 50 layer resnet"
    blocks = {18: [2, 2, 2, 2], 50: [3, 4, 6, 3]}[num_layers]
    block_type = {18: models.resnet.BasicBlock, 50: models.resnet.Bottleneck}[num_layers]
    model = ResNetMultiImageInput(block_type, blocks, in_channels=in_channels)

    if pretrained:
        print('loading imagnet weights on resnet...')
        loaded = model_zoo.load_url(models.resnet.model_urls['resnet{}'.format(num_layers)])
        # loaded['conv1.weight'] = torch.cat(
        #     (loaded['conv1.weight'], loaded['conv1.weight']), 1)
        # diff = model.load_state_dict(loaded, strict=False)
    return model


class ResnetEncoder(nn.Module):
    """Pytorch module for a resnet encoder
    """
    def __init__(self, num_layers, pretrained, in_channels=3):
        super(ResnetEncoder, self).__init__()

        self.num_ch_enc = np.array([64, 64, 128, 256, 512])

        resnets = {18: models.resnet18,
                   34: models.resnet34,
                   50: models.resnet50,
                   101: models.resnet101,
                   152: models.resnet152}

        if num_layers not in resnets:
            raise ValueError("{} is not a valid number of resnet layers".format(num_layers))

        if in_channels > 3:
            self.encoder = resnet_multiimage_input(num_layers, pretrained, in_channels)
        else:
            weights = models.ResNet101_Weights.IMAGENET1K_V1 if pretrained else None
            self.encoder = resnets[num_layers](weights=weights)

        if num_layers > 34:
            self.num_ch_enc[1:] *= 4

    def forward(self, x):
        self.features = []

        # input_image, normals = xx
        # x = (input_image - 0.45) / 0.225
        # x = torch.cat((input_image, normals),1)
        x = self.encoder.conv1(x)
        x = self.encoder.bn1(x)
        self.features.append(self.encoder.relu(x))
        self.features.append(self.encoder.layer1(self.encoder.maxpool(self.features[-1])))
        self.features.append(self.encoder.layer2(self.features[-1]))
        self.features.append(self.encoder.layer3(self.features[-1]))
        self.features.append(self.encoder.layer4(self.features[-1]))

        return self.features


class Decoder(nn.Module):
    def __init__(self, num_ch_enc, scales=range(4), num_output_channels=1, use_skips=True,
        kaiming_init=False, return_feats=False):
        super().__init__()

        self.num_output_channels = num_output_channels
        self.use_skips = use_skips
        self.upsample_mode = 'nearest'
        self.scales = scales

        self.return_feats = return_feats

        self.num_ch_enc = num_ch_enc
        self.num_ch_dec = np.array([16, 32, 64, 128, 256])

        # decoder
        self.convs = OrderedDict()
        for i in range(4, -1, -1):
            # upconv_0
            num_ch_in = self.num_ch_enc[-1] if i == 4 else self.num_ch_dec[i + 1]
            num_ch_out = self.num_ch_dec[i]
            self.convs[("upconv", i, 0)] = ConvBlock(num_ch_in, num_ch_out)

            # upconv_1
            num_ch_in = self.num_ch_dec[i]
            if self.use_skips and i > 0:
                num_ch_in += self.num_ch_enc[i - 1]
            num_ch_out = self.num_ch_dec[i]
            self.convs[("upconv", i, 1)] = ConvBlock(num_ch_in, num_ch_out)

        # for s in self.scales:
        self.convs[("dispconv", 0)] = Conv3x3(self.num_ch_dec[0], self.num_output_channels)

        self.decoder = nn.ModuleList(list(self.convs.values()))
        # self.sigmoid = nn.Sigmoid()

        if kaiming_init:
            print('init weights of decoder')
            for m in self.children():
                if isinstance(m, nn.Conv2d):
                    nn.init.kaiming_normal_(m.weight)
                    if m.bias is not None:
                        m.bias.data.fill_(0.01)

    def forward(self, input_features):
        x = input_features[-1]
        for i in range(4, -1, -1):
            x = self.convs[("upconv", i, 0)](x)
            x = [upsample(x)]
            if self.use_skips and i > 0:
                x += [input_features[i - 1]]
            x = torch.cat(x, 1)
            x = self.convs[("upconv", i, 1)](x)

        # assert self.scales[0] == 0
        final_conv = self.convs[("dispconv", 0)]
        out = final_conv(x)

        if self.return_feats:
            return out, input_features[-1]
        return out

class MultiHeadDecoder(nn.Module):
    def __init__(self, num_ch_enc, tasks, return_feats, use_skips):
        super().__init__()
        self.decoders = nn.ModuleDict({k:
            Decoder(num_ch_enc=num_ch_enc,
                    num_output_channels=num_ch,
                    scales=[0],
                    kaiming_init=False,
                    use_skips=use_skips,
                    return_feats=return_feats)
            for k, num_ch in tasks.items()})

    def forward(self, x):
        y = EasyDict({k: v(x) for k, v in self.decoders.items()})
        return y

class DenseMTL(nn.Module):
    def __init__(self, encoder, decoder):
        super().__init__()
        self.encoder = encoder
        self.decoder = decoder
    def forward(self, x):
        return self.decoder(self.encoder(x))