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
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
| import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Dict, List, Tuple, Optional
import numpy as np
class TDGHYOLOv7(nn.Module):
"""
TDGH-YOLOv7: Task-Driven Gaze Head for Cross-Driver Generalization
核心创新：
1. 多头注意力机制（空间、通道、交叉）
2. 任务驱动的特征学习
3. 跨域泛化损失
"""
def __init__(self,
num_classes: int = 0,
gaze_dim: int = 2,
pretrained: bool = True):
"""
初始化模型
Args:
num_classes: 分类类别数（可选）
gaze_dim: 视线维度（pitch, yaw）
pretrained: 是否使用预训练权重
"""
super().__init__()
self.backbone = self._build_backbone()
self.spatial_attention = SpatialAttentionHead(in_channels=256)
self.channel_attention = ChannelAttentionHead(in_channels=256)
self.cross_attention = CrossAttentionHead(in_channels=256)
self.feature_fusion = FeatureFusion(in_channels=256 * 3, out_channels=512)
self.gaze_head = nn.Sequential(
nn.Linear(512, 256),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(256, gaze_dim)
)
self.classification_head = None
if num_classes > 0:
self.classification_head = nn.Linear(512, num_classes)
def forward(self,
face_image: torch.Tensor,
eye_crop: Optional[torch.Tensor] = None) -> Dict:
"""
前向传播
Args:
face_image: 面部图像 (B, 3, H, W)
eye_crop: 眼睛裁剪图像 (B, 3, H', W')（可选）
Returns:
output: {
"gaze": (B, 2), # (pitch, yaw)
"class_logits": (B, num_classes)（可选）
}
"""
features = self.backbone(face_image)
spatial_features = self.spatial_attention(features)
channel_features = self.channel_attention(features)
cross_features = self.cross_attention(features)
fused_features = torch.cat([
spatial_features,
channel_features,
cross_features
], dim=1)
fused_features = self.feature_fusion(fused_features)
gaze = self.gaze_head(fused_features)
output = {"gaze": gaze}
if self.classification_head is not None:
output["class_logits"] = self.classification_head(fused_features)
return output
def _build_backbone(self) -> nn.Module:
"""构建骨干网络"""
return nn.Sequential(
nn.Conv2d(3, 32, 3, 2, 1),
nn.BatchNorm2d(32),
nn.SiLU(),
nn.Conv2d(32, 64, 3, 2, 1),
nn.BatchNorm2d(64),
nn.SiLU(),
nn.Conv2d(64, 128, 3, 2, 1),
nn.BatchNorm2d(128),
nn.SiLU(),
nn.Conv2d(128, 256, 3, 2, 1),
nn.BatchNorm2d(256),
nn.SiLU(),
nn.AdaptiveAvgPool2d(1),
nn.Flatten()
)
class SpatialAttentionHead(nn.Module):
"""空间注意力头"""
def __init__(self, in_channels: int):
super().__init__()
self.conv = nn.Conv2d(in_channels, 1, 1)
self.sigmoid = nn.Sigmoid()
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
Args:
x: (B, C, H, W)
Returns:
attended: (B, C, H, W)
"""
attention = self.conv(x)
attention = self.sigmoid(attention)
attended = x * attention
return attended
class ChannelAttentionHead(nn.Module):
"""通道注意力头"""
def __init__(self, in_channels: int, reduction: int = 16):
super().__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Sequential(
nn.Linear(in_channels, in_channels // reduction),
nn.ReLU(),
nn.Linear(in_channels // reduction, in_channels),
nn.Sigmoid()
)
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
Args:
x: (B, C, H, W)
Returns:
attended: (B, C, H, W)
"""
b, c, h, w = x.shape
y = self.avg_pool(x).view(b, c)
y = self.fc(y).view(b, c, 1, 1)
attended = x * y
return attended
class CrossAttentionHead(nn.Module):
"""交叉注意力头"""
def __init__(self, in_channels: int):
super().__init__()
self.query_conv = nn.Conv2d(in_channels, in_channels // 8, 1)
self.key_conv = nn.Conv2d(in_channels, in_channels // 8, 1)
self.value_conv = nn.Conv2d(in_channels, in_channels, 1)
self.gamma = nn.Parameter(torch.zeros(1))
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
Args:
x: (B, C, H, W)
Returns:
attended: (B, C, H, W)
"""
b, c, h, w = x.shape
query = self.query_conv(x).view(b, -1, h * w).permute(0, 2, 1)
key = self.key_conv(x).view(b, -1, h * w)
value = self.value_conv(x).view(b, -1, h * w)
attention = torch.bmm(query, key)
attention = F.softmax(attention, dim=-1)
out = torch.bmm(value, attention.permute(0, 2, 1))
out = out.view(b, c, h, w)
attended = self.gamma * out + x
return attended
class FeatureFusion(nn.Module):
"""特征融合"""
def __init__(self, in_channels: int, out_channels: int):
super().__init__()
self.fc = nn.Sequential(
nn.Linear(in_channels, out_channels),
nn.ReLU(),
nn.Dropout(0.3)
)
def forward(self, x: torch.Tensor) -> torch.Tensor:
return self.fc(x)
if __name__ == "__main__":
model = TDGHYOLOv7(gaze_dim=2)
face_image = torch.randn(1, 3, 224, 224)
output = model(face_image)
print(f"视线预测: {output['gaze']}")
print(f"模型参数: {sum(p.numel() for p in model.parameters()) / 1e6:.2f}M")
|