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
|
"""Util functions for efficient net."""
from functools import partial
import math
from typing import Any, Optional, Tuple, Type
from omegaconf import OmegaConf
import torch
from torch import nn, Tensor
import torch.functional as F
def calculate_output_image_size(
image_size: Optional[Tuple[int, int]], stride: int
) -> Optional[Tuple[int, int]]:
"""Calculates the output image size when using conv2d with same padding."""
if image_size is None:
return None
height = int(math.ceil(image_size[0] / stride))
width = int(math.ceil(image_size[1] / stride))
return height, width
def drop_connection(x: Tensor, p: float, training: bool) -> Tensor:
"""Drop connection.
Drops the entire convolution with a given survival probability.
Args:
x (Tensor): Input tensor.
p (float): Survival probability between 0.0 and 1.0.
training (bool): The running mode.
Shapes:
- x: :math: `(B, C, W, H)`.
- out: :math: `(B, C, W, H)`.
where B is the batch size, C is the number of channels, W is the width, and H
is the height.
Returns:
out (Tensor): Output after drop connection.
"""
assert 0.0 <= p <= 1.0, "p must be in range of [0, 1]"
if not training:
return x
bsz = x.shape[0]
survival_prob = 1 - p
# Generate a binary tensor mask according to probability (p for 0, 1-p for 1)
random_tensor = survival_prob
random_tensor += torch.rand([bsz, 1, 1, 1]).type_as(x)
binary_tensor = torch.floor(random_tensor)
out = x / survival_prob * binary_tensor
return out
def get_same_padding_conv2d(image_size: Optional[Tuple[int, int]]) -> Type[nn.Conv2d]:
if image_size is None:
return Conv2dDynamicSamePadding
return partial(Conv2dStaticSamePadding, image_size=image_size)
class Conv2dDynamicSamePadding(nn.Conv2d):
def __init__(
self,
in_channels: int,
out_channels: int,
kernel_size: int,
stride: int = 1,
dilation: int = 1,
groups: int = 1,
bias: bool = True,
) -> None:
super().__init__(
in_channels, out_channels, kernel_size, stride, 0, dilation, groups, bias
)
self.stride = [self.stride] * 2
def forward(self, x: Tensor) -> Tensor:
ih, iw = x.shape[-2:]
kh, kw = self.weight.shape[-2:]
sh, sw = self.stride
oh, ow = math.ceil(ih / sh), math.ceil(iw / sw)
pad_h = max((oh - 1) * self.stride[0] + (kh - 1) * self.dilation[0] + 1 - ih, 0)
pad_w = max((ow - 1) * self.stride[1] + (kw - 1) * self.dilation[1] + 1 - iw, 0)
if pad_h > 0 or pad_w > 0:
x = F.pad(
x, [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2]
)
return F.conv2d(
x,
self.weight,
self.bias,
self.stride,
self.padding,
self.dilation,
self.groups,
)
class Conv2dStaticSamePadding(nn.Conv2d):
def __init__(
self,
in_channels: int,
out_channels: int,
kernel_size: int,
image_size: Tuple[int, int],
stride: int = 1,
**kwargs: Any
):
super().__init__(in_channels, out_channels, kernel_size, stride, **kwargs)
self.stride = [self.stride] * 2
# Calculate padding based on image size and save it.
ih, iw = image_size
kh, kw = self.weight.shape[-2:]
sh, sw = self.stride
oh, ow = math.ceil(ih / sh), math.ceil(iw / sw)
pad_h = max((oh - 1) * self.stride[0] + (kh - 1) * self.dilation[0] + 1 - ih, 0)
pad_w = max((ow - 1) * self.stride[1] + (kw - 1) * self.dilation[1] + 1 - iw, 0)
if pad_h > 0 or pad_w > 0:
self.static_padding = nn.ZeroPad2d(
(pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2)
)
else:
self.static_padding = nn.Identity()
def forward(self, x: Tensor) -> Tensor:
x = self.static_padding(x)
x = F.pad(
x,
self.weight,
self.bias,
self.stride,
self.padding,
self.dilation,
self.groups,
)
return x
def round_filters(filters: int, arch: Tuple[float, float, float]) -> int:
multiplier = arch[0]
divisor = 8
filters *= multiplier
new_filters = max(divisor, (filters + divisor // 2) // divisor * divisor)
if new_filters < 0.9 * filters:
new_filters += divisor
return int(new_filters)
def round_repeats(repeats: int, arch: Tuple[float, float, float]) -> int:
return int(math.ceil(arch[1] * repeats))
def block_args():
keys = [
"num_repeats",
"kernel_size",
"strides",
"expand_ratio",
"in_channels",
"out_channels",
"se_ratio",
]
args = [
[1, 3, (1, 1), 1, 32, 16, 0.25],
[2, 3, (2, 2), 6, 16, 24, 0.25],
[2, 5, (2, 2), 6, 24, 40, 0.25],
[3, 3, (2, 2), 6, 40, 80, 0.25],
[3, 5, (1, 1), 6, 80, 112, 0.25],
[4, 5, (2, 2), 6, 112, 192, 0.25],
[1, 3, (1, 1), 6, 192, 320, 0.25],
]
block_args_ = []
for row in args:
block_args_.append(OmegaConf.create(dict(zip(keys, row))))
return block_args_
|
