File size: 13,164 Bytes
8eab354
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
"""
Feature extraction module for signature verification using CNN-based approaches.
"""

import torch
import torch.nn as nn
import torchvision.models as models
from typing import Tuple, Optional
import torch.nn.functional as F


class SignatureFeatureExtractor(nn.Module):
    """
    CNN-based feature extractor for signature images.
    """
    
    def __init__(self, 
                 backbone: str = 'resnet18',
                 feature_dim: int = 512,
                 pretrained: bool = True,
                 freeze_backbone: bool = False):
        """
        Initialize the feature extractor.
        
        Args:
            backbone: Backbone architecture ('resnet18', 'resnet34', 'resnet50', 'efficientnet')
            feature_dim: Dimension of output features
            pretrained: Whether to use pretrained weights
            freeze_backbone: Whether to freeze backbone parameters
        """
        super(SignatureFeatureExtractor, self).__init__()
        
        self.backbone_name = backbone
        self.feature_dim = feature_dim
        self.pretrained = pretrained
        
        # Load backbone
        self.backbone = self._get_backbone(backbone, pretrained)
        
        # Freeze backbone if specified
        if freeze_backbone:
            for param in self.backbone.parameters():
                param.requires_grad = False
        
        # Get the number of input features from backbone
        if 'resnet' in backbone:
            backbone_features = self.backbone.fc.in_features
            self.backbone.fc = nn.Identity()  # Remove final classification layer
        elif 'efficientnet' in backbone:
            backbone_features = self.backbone.classifier.in_features
            self.backbone.classifier = nn.Identity()
        else:
            raise ValueError(f"Unsupported backbone: {backbone}")
        
        # Feature projection layers
        self.feature_projection = nn.Sequential(
            nn.Linear(backbone_features, feature_dim * 2),
            nn.BatchNorm1d(feature_dim * 2),
            nn.ReLU(inplace=True),
            nn.Dropout(0.3),
            nn.Linear(feature_dim * 2, feature_dim),
            nn.BatchNorm1d(feature_dim),
            nn.ReLU(inplace=True)
        )
        
        # Initialize weights
        self._initialize_weights()
    
    def _get_backbone(self, backbone: str, pretrained: bool):
        """Get the backbone model."""
        if backbone == 'resnet18':
            return models.resnet18(pretrained=pretrained)
        elif backbone == 'resnet34':
            return models.resnet34(pretrained=pretrained)
        elif backbone == 'resnet50':
            return models.resnet50(pretrained=pretrained)
        elif backbone == 'efficientnet_b0':
            return models.efficientnet_b0(pretrained=pretrained)
        elif backbone == 'efficientnet_b1':
            return models.efficientnet_b1(pretrained=pretrained)
        else:
            raise ValueError(f"Unsupported backbone: {backbone}")
    
    def _initialize_weights(self):
        """Initialize weights for the projection layers."""
        for m in self.feature_projection.modules():
            if isinstance(m, nn.Linear):
                nn.init.xavier_uniform_(m.weight)
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.BatchNorm1d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Forward pass through the feature extractor.
        
        Args:
            x: Input signature images (B, C, H, W)
            
        Returns:
            Extracted features (B, feature_dim)
        """
        # Extract features using backbone
        features = self.backbone(x)
        
        # Project to desired feature dimension
        features = self.feature_projection(features)
        
        # L2 normalize features
        features = F.normalize(features, p=2, dim=1)
        
        return features


class CustomCNNFeatureExtractor(nn.Module):
    """
    Custom CNN architecture specifically designed for signature verification.
    """
    
    def __init__(self, 
                 input_channels: int = 3,
                 feature_dim: int = 512,
                 dropout_rate: float = 0.3):
        """
        Initialize the custom CNN feature extractor.
        
        Args:
            input_channels: Number of input channels
            feature_dim: Dimension of output features
            dropout_rate: Dropout rate for regularization
        """
        super(CustomCNNFeatureExtractor, self).__init__()
        
        self.feature_dim = feature_dim
        self.dropout_rate = dropout_rate
        
        # Convolutional layers
        self.conv_layers = nn.Sequential(
            # First block
            nn.Conv2d(input_channels, 32, kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),
            nn.Conv2d(32, 32, kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(2, 2),
            
            # Second block
            nn.Conv2d(32, 64, kernel_size=3, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, 64, kernel_size=3, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(2, 2),
            
            # Third block
            nn.Conv2d(64, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(2, 2),
            
            # Fourth block
            nn.Conv2d(128, 256, kernel_size=3, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 256, kernel_size=3, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(2, 2),
            
            # Fifth block
            nn.Conv2d(256, 512, kernel_size=3, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512, kernel_size=3, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.AdaptiveAvgPool2d((1, 1))
        )
        
        # Fully connected layers
        self.fc_layers = nn.Sequential(
            nn.Flatten(),
            nn.Linear(512, feature_dim * 2),
            nn.BatchNorm1d(feature_dim * 2),
            nn.ReLU(inplace=True),
            nn.Dropout(dropout_rate),
            nn.Linear(feature_dim * 2, feature_dim),
            nn.BatchNorm1d(feature_dim),
            nn.ReLU(inplace=True)
        )
        
        # Initialize weights
        self._initialize_weights()
    
    def _initialize_weights(self):
        """Initialize weights for all layers."""
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.xavier_uniform_(m.weight)
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Forward pass through the custom CNN.
        
        Args:
            x: Input signature images (B, C, H, W)
            
        Returns:
            Extracted features (B, feature_dim)
        """
        # Extract features using convolutional layers
        features = self.conv_layers(x)
        
        # Project to desired feature dimension
        features = self.fc_layers(features)
        
        # L2 normalize features
        features = F.normalize(features, p=2, dim=1)
        
        return features


class MultiScaleFeatureExtractor(nn.Module):
    """
    Multi-scale feature extractor that captures features at different scales.
    """
    
    def __init__(self, 
                 input_channels: int = 3,
                 feature_dim: int = 512,
                 scales: list = [1, 2, 4]):
        """
        Initialize the multi-scale feature extractor.
        
        Args:
            input_channels: Number of input channels
            feature_dim: Dimension of output features
            scales: List of scales for multi-scale processing
        """
        super(MultiScaleFeatureExtractor, self).__init__()
        
        self.scales = scales
        self.feature_dim = feature_dim
        
        # Create feature extractors for each scale
        self.scale_extractors = nn.ModuleList()
        for scale in scales:
            extractor = CustomCNNFeatureExtractor(
                input_channels=input_channels,
                feature_dim=feature_dim // len(scales)
            )
            self.scale_extractors.append(extractor)
        
        # Fusion layer
        self.fusion = nn.Sequential(
            nn.Linear(feature_dim, feature_dim),
            nn.BatchNorm1d(feature_dim),
            nn.ReLU(inplace=True),
            nn.Dropout(0.3)
        )
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Forward pass through the multi-scale extractor.
        
        Args:
            x: Input signature images (B, C, H, W)
            
        Returns:
            Multi-scale features (B, feature_dim)
        """
        scale_features = []
        
        for i, scale in enumerate(self.scales):
            # Resize input to different scales
            if scale != 1:
                scaled_x = F.interpolate(x, scale_factor=1/scale, mode='bilinear', align_corners=False)
            else:
                scaled_x = x
            
            # Extract features at this scale
            features = self.scale_extractors[i](scaled_x)
            scale_features.append(features)
        
        # Concatenate features from all scales
        multi_scale_features = torch.cat(scale_features, dim=1)
        
        # Fuse features
        fused_features = self.fusion(multi_scale_features)
        
        # L2 normalize features
        fused_features = F.normalize(fused_features, p=2, dim=1)
        
        return fused_features


class AttentionFeatureExtractor(nn.Module):
    """
    Feature extractor with attention mechanism for focusing on important signature regions.
    """
    
    def __init__(self, 
                 input_channels: int = 3,
                 feature_dim: int = 512,
                 attention_dim: int = 256):
        """
        Initialize the attention-based feature extractor.
        
        Args:
            input_channels: Number of input channels
            feature_dim: Dimension of output features
            attention_dim: Dimension of attention features
        """
        super(AttentionFeatureExtractor, self).__init__()
        
        self.feature_dim = feature_dim
        self.attention_dim = attention_dim
        
        # Base feature extractor
        self.base_extractor = CustomCNNFeatureExtractor(
            input_channels=input_channels,
            feature_dim=feature_dim
        )
        
        # Attention mechanism
        self.attention_conv = nn.Sequential(
            nn.Conv2d(512, attention_dim, kernel_size=1),
            nn.BatchNorm2d(attention_dim),
            nn.ReLU(inplace=True),
            nn.Conv2d(attention_dim, 1, kernel_size=1),
            nn.Sigmoid()
        )
        
        # Feature refinement
        self.feature_refinement = nn.Sequential(
            nn.Linear(feature_dim, feature_dim),
            nn.BatchNorm1d(feature_dim),
            nn.ReLU(inplace=True),
            nn.Dropout(0.3)
        )
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Forward pass through the attention-based extractor.
        
        Args:
            x: Input signature images (B, C, H, W)
            
        Returns:
            Attention-weighted features (B, feature_dim)
        """
        # Get base features
        base_features = self.base_extractor(x)
        
        # Get attention map (simplified - in practice, you'd extract intermediate features)
        # For now, we'll use a simplified approach
        attention_map = self.attention_conv(x.mean(dim=1, keepdim=True))
        
        # Apply attention to features (simplified)
        attended_features = base_features * attention_map.mean(dim=[2, 3], keepdim=True).squeeze()
        
        # Refine features
        refined_features = self.feature_refinement(attended_features)
        
        # L2 normalize features
        refined_features = F.normalize(refined_features, p=2, dim=1)
        
        return refined_features