-
Notifications
You must be signed in to change notification settings - Fork 0
/
architecture.py
184 lines (155 loc) · 7.11 KB
/
architecture.py
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
"""
Architecture of the Neural Network.
Modified version of the DenseNet PyTorch implementation presented in
https://amaarora.github.io/2020/08/02/densenets.html (original paper:
https://arxiv.org/abs/1608.06993).
Modifications:
- increased input channels from 3 to 5
- decreased growth rate from 64 to 12 (similar performance, less parameters)
- added custom projection head to use flags (camera/month/etc)
@version 2022-12
@author Raúl Coterillo
"""
from __future__ import annotations
import torch.nn.functional as F
import torch.nn as nn
import torch
class _Transition(nn.Sequential):
""" Transition Layer. """
def __init__(self, num_input_features, num_output_features):
super(_Transition, self).__init__()
self.add_module('norm', nn.BatchNorm2d(num_input_features))
self.add_module('relu', nn.ReLU(inplace=True))
self.add_module('conv', nn.Conv2d(num_input_features, num_output_features,
kernel_size=1, stride=1, bias=False))
self.add_module('pool', nn.AvgPool2d(kernel_size=2, stride=2))
class _DenseLayer(nn.Module):
""" Dense Layer. """
def __init__(self, num_input_features, growth_rate, bn_size, drop_rate):
super(_DenseLayer, self).__init__()
self.add_module('norm1', nn.BatchNorm2d(num_input_features)),
self.add_module('relu1', nn.ReLU(inplace=True)),
self.add_module('conv1', nn.Conv2d(num_input_features, bn_size *
growth_rate, kernel_size=1, stride=1,
bias=False)),
self.add_module('norm2', nn.BatchNorm2d(bn_size * growth_rate)),
self.add_module('relu2', nn.ReLU(inplace=True)),
self.add_module('conv2', nn.Conv2d(bn_size * growth_rate, growth_rate,
kernel_size=3, stride=1, padding=1,
bias=False)),
self.drop_rate = float(drop_rate)
def bn_function(self, inputs):
""" Bottleneck function. """
concated_features = torch.cat(inputs, 1)
bottleneck_output = self.conv1(self.relu1(self.norm1(concated_features)))
return bottleneck_output
def forward(self, input):
if isinstance(input, torch.Tensor):
prev_features = [input]
else:
prev_features = input
bottleneck_output = self.bn_function(prev_features)
new_features = self.conv2(self.relu2(self.norm2(bottleneck_output)))
if self.drop_rate > 0:
new_features = F.dropout(new_features, p=self.drop_rate,
training=self.training)
return new_features
class _DenseBlock(nn.ModuleDict):
""" Dense Block. """
def __init__(self, num_layers, num_input_features, bn_size, growth_rate, drop_rate):
super(_DenseBlock, self).__init__()
for i in range(num_layers):
layer = _DenseLayer(
num_input_features + i * growth_rate,
growth_rate=growth_rate,
bn_size=bn_size,
drop_rate=drop_rate
)
self.add_module('denselayer%d' % (i + 1), layer)
def forward(self, init_features):
features = [init_features]
for name, layer in self.items():
new_features = layer(features)
features.append(new_features)
return torch.cat(features, 1)
class DenseNet(nn.Module):
def __init__(self,
num_classes: int = 3,
growth_rate: int = 12,
block_config: list[int] = (6, 12, 24, 16),
num_init_features: int =64,
bn_size: int = 4,
drop_rate: float = 0,
flags_size: int = None
) -> None:
super(DenseNet, self).__init__()
# Convolution and pooling part from table-1
self.features = nn.Sequential()
self.features.add_module('conv0', nn.Conv2d(5, num_init_features, kernel_size=7, stride=2, padding=3, bias=False))
self.features.add_module('norm0', nn.BatchNorm2d(num_init_features))
self.features.add_module('relu0', nn.ReLU(inplace=True))
self.features.add_module('pool0', nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
# Add multiple denseblocks based on config
# for densenet-121 config: [6,12,24,16]
num_features = num_init_features
for i, num_layers in enumerate(block_config):
block = _DenseBlock(
num_layers=num_layers,
num_input_features=num_features,
bn_size=bn_size,
growth_rate=growth_rate,
drop_rate=drop_rate
)
self.features.add_module('denseblock%d' % (i + 1), block)
num_features = num_features + num_layers * growth_rate
if i != len(block_config) - 1:
# add transition layer between denseblocks to downsample
trans = _Transition(num_input_features=num_features,
num_output_features=num_features // 2)
self.features.add_module('transition%d' % (i + 1), trans)
num_features = num_features // 2
# Final batch norm
self.features.add_module('norm5', nn.BatchNorm2d(num_features))
# Linear layers
self.num_classes = num_classes
self.features_size = num_features
self.flags_size = flags_size if flags_size is not None else 0
fts, fgs = self.features_size, self.flags_size
if self.flags_size == 0:
self.flags_size = None
self.lin1 = nn.Linear(in_features=fts+fgs, out_features=fts//2)
self.lin2 = nn.Sequential()
self.lin2.add_module("lin0", nn.Linear(in_features=fts//2 + fgs, out_features=num_classes))
self.lin2.add_module("sig0", nn.Sigmoid())
self.flags_size
# Official weight initizalization from torch repo.
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.constant_(m.bias, 0)
def forward(self, x):
# if no flags are used
if self.flags_size is None:
out = self.features(x.to(torch.float32))
out = F.relu(out, inplace=True)
out = F.adaptive_avg_pool2d(out, (1, 1))
out = torch.flatten(out, 1)
out = self.lin2(self.lin1(out))
# if flags are used
else:
image, flags = x[0], x[1]
image: torch.Tensor = image.to(torch.float32)
flags: torch.Tensor = flags.to(torch.float32)
out = self.features(image)
out = F.relu(out, inplace=True)
out = F.adaptive_avg_pool2d(out, (1, 1))
out = torch.flatten(out, 1)
out = torch.cat((out, flags), 1)
out = self.lin1(out)
out = torch.cat((out, flags), 1)
out = self.lin2(out)
return out