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
| """
论文复现：LDIE-FDNet完整实现
论文：LDIE-FDNet: Lightweight Dynamic Image Enhancement-Enabled Real-time Fatigue Driving Detection Network
作者：Dong, Zhang, Liu
期刊：PLOS ONE, 2026
DOI: 10.1371/journal.pone.0346055
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
import cv2
import numpy as np
from typing import List, Dict, Tuple
class LDIEFDNet(nn.Module):
"""
LDIE-FDNet完整网络
包含：
- MSR-LIENET (低光照增强)
- GSConv_C3k2 Backbone
- DHFAR-Net Neck
- PIoU Loss
"""
def __init__(self, num_classes: int = 5):
"""
Args:
num_classes: 检测类别数
0: normal, 1: eye_closed, 2: yawn, 3: smoking, 4: phone
"""
super().__init__()
self.msr_lienet = MSRLIENET()
self.backbone = nn.ModuleList([
nn.Sequential(
nn.Conv2d(3, 16, 3, stride=2, padding=1, bias=False),
nn.BatchNorm2d(16),
nn.SiLU(inplace=True)
),
GSConv_C3k2(16, 32, n=1),
GSConv_C3k2(32, 64, n=2),
GSConv_C3k2(64, 128, n=2),
GSConv_C3k2(128, 256, n=1),
nn.Sequential(
nn.Conv2d(256, 256, 1, bias=False),
nn.BatchNorm2d(256),
nn.SiLU(inplace=True),
nn.MaxPool2d(5, stride=1, padding=2),
nn.MaxPool2d(5, stride=1, padding=2),
nn.MaxPool2d(5, stride=1, padding=2),
)
])
self.neck = DHFARNet([64, 128, 256])
self.head = nn.Conv2d(256, num_classes + 5, 1)
self.piou_loss = PIoULoss()
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
Args:
x: 输入图像 (B, 3, 640, 640)
Returns:
out: 检测输出 (B, num_classes+5, H, W)
"""
x = self.msr_lienet(x)
features = []
for i, layer in enumerate(self.backbone):
x = layer(x)
if i in [2, 3, 4]:
features.append(x)
x = self.neck(features + [x])
out = self.head(x)
return out
class FatigueDetector:
"""疲劳检测完整管道"""
def __init__(self, model_path: str, device: str = 'cuda'):
self.device = device
self.model = LDIEFDNet(num_classes=5)
if model_path:
self.model.load_state_dict(torch.load(model_path, map_location=device))
self.model.to(device)
self.model.eval()
self.decision = FatigueDecision(fps=20)
self.class_names = ['normal', 'eye_closed', 'yawn', 'smoking', 'phone']
def detect_frame(self, frame: np.ndarray) -> Dict:
"""
检测单帧
Args:
frame: BGR图像 (H, W, 3)
Returns:
result: 检测结果
"""
image = cv2.resize(frame, (640, 640))
image = image[:, :, ::-1]
image = image.transpose(2, 0, 1)
image = torch.from_numpy(image).float() / 255.0
image = image.unsqueeze(0).to(self.device)
with torch.no_grad():
output = self.model(image)
detections = self.postprocess(output)
decision_result = self.decision.update(detections)
return {
'detections': detections,
'fatigue': decision_result
}
def postprocess(self, output: torch.Tensor, conf_thresh: float = 0.5) -> List[Dict]:
"""后处理：提取检测框"""
detections = []
output = output.squeeze(0)
obj_conf = output[4].sigmoid()
mask = obj_conf > conf_thresh
if mask.any():
class_conf, class_idx = output[5:].sigmoid().max(dim=0)
class_conf = class_conf[mask]
class_idx = class_idx[mask]
for conf, idx in zip(class_conf, class_idx):
detections.append({
'class': self.class_names[idx.item()],
'conf': conf.item()
})
return detections
if __name__ == "__main__":
model = LDIEFDNet(num_classes=5)
x = torch.randn(1, 3, 640, 640)
output = model(x)
print(f"输入形状: {x.shape}")
print(f"输出形状: {output.shape}")
print(f"参数量: {sum(p.numel() for p in model.parameters()):,}")
print(f"计算量: {sum(p.numel() * p.numel() for p in model.parameters()) / 1e9:.2f} GFLOPs")
|