-
Notifications
You must be signed in to change notification settings - Fork 153
/
backbone.py
205 lines (169 loc) · 6.62 KB
/
backbone.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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
# Modified from ultralytics/yolov5 by Zhiqiang Wang
from pathlib import Path
from collections import OrderedDict
import yaml
import torch
from torch import nn, Tensor
from typing import List, Dict, Optional
from .common import Conv, Bottleneck, SPP, DWConv, Focus, BottleneckCSP, Concat
from .experimental import MixConv2d, CrossConv, C3
from .box_head import Detect
class YoloBackbone(nn.Module):
def __init__(
self,
yolo_body: nn.Module,
return_layers: dict,
out_channels: List[int],
):
super().__init__()
self.body = IntermediateLayerGetter(
yolo_body.model,
return_layers=return_layers,
save_list=yolo_body.save_list,
)
self.out_channels = out_channels
def forward(self, x: Tensor):
x = self.body(x)
out: List[Tensor] = []
for name, feature in x.items():
out.append(feature)
return out
class YoloBody(nn.Module):
__annotations__ = {
"save_list": List[int],
}
def __init__(self, layers, save_list):
super().__init__()
# Define model
self.model = nn.Sequential(*layers)
self.save_list = save_list
# Init weights, biases
self._initialize_weights()
def forward(self, x: Tensor) -> Tensor:
out = x
y: List[Tensor] = []
for i, m in enumerate(self.model):
if m.f > 0: # Concat layer
out = torch.cat([out, y[sorted(self.save_list).index(m.f)]], 1)
else:
out = m(out) # run
if i in self.save_list:
y.append(out) # save output
return out
def _initialize_weights(self) -> None:
for m in self.modules():
if isinstance(m, nn.Conv2d):
pass # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, nn.BatchNorm2d):
m.eps = 1e-3
m.momentum = 0.03
elif isinstance(m, (nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6)):
m.inplace = True
def parse_model(model_dict, in_channels=3):
head_info = ()
anchors, num_classes = model_dict['anchors'], model_dict['nc']
num_anchors = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors
num_outputs = num_anchors * (num_classes + 5)
c2 = in_channels
layers, save_list, channels = [], [], [c2] # layers, save list, channels out
# from, number, module, args
for i, (f, n, m, args) in enumerate(model_dict['backbone'] + model_dict['head']):
m = eval(m) if isinstance(m, str) else m # eval strings
for j, a in enumerate(args):
try:
args[j] = eval(a) if isinstance(a, str) else a # eval strings
except NameError:
pass
n = max(round(n * model_dict['depth_multiple']), 1) if n > 1 else n # depth gain
if m in [Conv, Bottleneck, SPP, DWConv, MixConv2d, Focus, CrossConv, BottleneckCSP, C3]:
c1, c2 = channels[f], args[0]
c2 = _make_divisible(c2 * model_dict['width_multiple'], 8) if c2 != num_outputs else c2
args = [c1, c2, *args[1:]]
if m in [BottleneckCSP, C3]:
args.insert(2, n)
n = 1
elif m is nn.BatchNorm2d:
args = [channels[f]]
elif m is Concat:
c2 = sum([channels[-1 if x == -1 else x + 1] for x in f])
elif m is Detect:
num_layers, anchor_grids = f, args[-1]
out_channels = [channels[x + 1] for x in f]
head_info = (out_channels, anchor_grids, num_layers)
continue
else:
c2 = channels[f]
module = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args)
module.f = -1 if f == -1 else f[-1]
save_list.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)
layers.append(module)
channels.append(c2)
return layers, save_list, head_info
def _make_divisible(v: float, divisor: int, min_value: Optional[int] = None) -> int:
"""
This function is taken from the original tf repo.
It ensures that all layers have a channel number that is divisible by 8
It can be seen here:
https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
:param v:
:param divisor:
:param min_value:
:return:
"""
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
# Make sure that round down does not go down by more than 10%.
if new_v < 0.9 * v:
new_v += divisor
return new_v
class IntermediateLayerGetter(nn.ModuleDict):
"""
Module wrapper that returns intermediate layers from a model
"""
_version = 2
__annotations__ = {
"return_layers": Dict[str, str],
"save_list": List[int],
}
def __init__(self, model, return_layers, save_list):
if not set(return_layers).issubset([name for name, _ in model.named_children()]):
raise ValueError("return_layers are not present in model")
orig_return_layers = return_layers
return_layers = {str(k): str(v) for k, v in return_layers.items()}
layers = OrderedDict()
for name, module in model.named_children():
layers[name] = module
if name in return_layers:
del return_layers[name]
if not return_layers:
break
super().__init__(layers)
self.return_layers = orig_return_layers
self.save_list = save_list
def forward(self, x):
out = OrderedDict()
y: List[Tensor] = []
for i, (name, module) in enumerate(self.items()):
if module.f > 0: # Concat layer
x = torch.cat([x, y[sorted(self.save_list).index(module.f)]], 1)
else:
x = module(x) # run
if i in self.save_list:
y.append(x) # save output
if name in self.return_layers:
out_name = self.return_layers[name]
out[out_name] = x
return out
def darknet(cfg_path='yolov5s.yaml'):
cfg_path = Path(__file__).parent.absolute().joinpath(cfg_path)
with open(cfg_path) as f:
model_dict = yaml.load(f, Loader=yaml.FullLoader)
layers, save_list, head_info = parse_model(model_dict, in_channels=3)
body = YoloBody(layers, save_list)
backbone = YoloBackbone(
yolo_body=body,
return_layers={str(key): str(i) for i, key in enumerate(head_info[2])},
out_channels=head_info[0],
)
return backbone, head_info[1]