File size: 17,386 Bytes
faaaf38
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
"""
Deepfake Hunter - Model Management Module

Handles loading, caching, and managing pre-trained models for deepfake detection.

Features:
- Automatic model downloading from Hugging Face Hub
- Model caching for performance
- GPU acceleration with CPU fallback
- Model ensemble voting
- Version management

Author: Deepfake Hunter Team
License: MIT
"""

import warnings
warnings.filterwarnings('ignore')

from typing import Dict, List, Optional, Any, Union
from pathlib import Path
from dataclasses import dataclass
import hashlib
import json
import os

import torch
import torch.nn as nn
from torchvision import models
import numpy as np
from loguru import logger
from huggingface_hub import hf_hub_download, list_repo_files
from tqdm import tqdm


@dataclass
class ModelConfig:
    """Configuration for a pre-trained model"""
    name: str
    repo_id: str
    filename: str
    model_type: str  # 'efficientnet', 'resnet', '3dcnn', etc.
    input_size: tuple
    version: str
    checksum: str = ""
    url: Optional[str] = None


class ModelCache:
    """
    Manages model caching to avoid re-downloading and re-loading

    Models are cached in memory and on disk.
    """

    def __init__(self, cache_dir: Optional[Path] = None):
        if cache_dir is None:
            cache_dir = Path.home() / ".cache" / "deepfake-hunter" / "models"

        self.cache_dir = Path(cache_dir)
        self.cache_dir.mkdir(parents=True, exist_ok=True)

        self._memory_cache: Dict[str, nn.Module] = {}
        self._config_cache: Dict[str, ModelConfig] = {}

        logger.info(f"ModelCache initialized at: {self.cache_dir}")

    def get_cache_path(self, model_name: str) -> Path:
        """Get path for cached model file"""
        return self.cache_dir / f"{model_name}.pth"

    def is_cached(self, model_name: str) -> bool:
        """Check if model is cached on disk"""
        return self.get_cache_path(model_name).exists()

    def load_from_cache(self, model_name: str, device: str = "cpu") -> Optional[nn.Module]:
        """Load model from memory or disk cache"""
        # Check memory cache first
        if model_name in self._memory_cache:
            logger.info(f"Loading {model_name} from memory cache")
            return self._memory_cache[model_name]

        # Check disk cache
        cache_path = self.get_cache_path(model_name)
        if cache_path.exists():
            try:
                logger.info(f"Loading {model_name} from disk cache")
                model = torch.load(cache_path, map_location=device)
                self._memory_cache[model_name] = model
                return model
            except Exception as e:
                logger.error(f"Failed to load cached model {model_name}: {e}")
                return None

        return None

    def save_to_cache(self, model_name: str, model: nn.Module):
        """Save model to memory and disk cache"""
        try:
            # Save to memory
            self._memory_cache[model_name] = model

            # Save to disk
            cache_path = self.get_cache_path(model_name)
            torch.save(model, cache_path)
            logger.info(f"Saved {model_name} to cache")
        except Exception as e:
            logger.error(f"Failed to cache model {model_name}: {e}")

    def clear_cache(self, model_name: Optional[str] = None):
        """Clear cache for specific model or all models"""
        if model_name:
            # Clear specific model
            if model_name in self._memory_cache:
                del self._memory_cache[model_name]

            cache_path = self.get_cache_path(model_name)
            if cache_path.exists():
                cache_path.unlink()

            logger.info(f"Cleared cache for {model_name}")
        else:
            # Clear all
            self._memory_cache.clear()

            for cache_file in self.cache_dir.glob("*.pth"):
                cache_file.unlink()

            logger.info("Cleared all model caches")


class EfficientNetDetector(nn.Module):
    """
    EfficientNet-based spatial artifact detector

    Fine-tuned on FaceForensics++ dataset for deepfake detection
    """

    def __init__(self, num_classes: int = 2, pretrained: bool = True):
        super().__init__()

        # Load EfficientNet-B4 (good balance of speed and accuracy)
        self.backbone = models.efficientnet_b4(pretrained=pretrained)

        # Replace classifier
        in_features = self.backbone.classifier[1].in_features
        self.backbone.classifier = nn.Sequential(
            nn.Dropout(p=0.4, inplace=True),
            nn.Linear(in_features, num_classes)
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return self.backbone(x)

    def predict_proba(self, x: torch.Tensor) -> torch.Tensor:
        """Get probabilities instead of logits"""
        logits = self.forward(x)
        return torch.softmax(logits, dim=1)


class CNN3DTemporalDetector(nn.Module):
    """
    3D CNN for temporal inconsistency detection

    Analyzes sequences of frames to detect unnatural temporal patterns
    """

    def __init__(self, num_classes: int = 2, input_channels: int = 3):
        super().__init__()

        # 3D convolutional layers
        self.conv1 = nn.Conv3d(input_channels, 64, kernel_size=(3, 3, 3), padding=(1, 1, 1))
        self.bn1 = nn.BatchNorm3d(64)
        self.pool1 = nn.MaxPool3d(kernel_size=(1, 2, 2))

        self.conv2 = nn.Conv3d(64, 128, kernel_size=(3, 3, 3), padding=(1, 1, 1))
        self.bn2 = nn.BatchNorm3d(128)
        self.pool2 = nn.MaxPool3d(kernel_size=(2, 2, 2))

        self.conv3 = nn.Conv3d(128, 256, kernel_size=(3, 3, 3), padding=(1, 1, 1))
        self.bn3 = nn.BatchNorm3d(256)
        self.pool3 = nn.MaxPool3d(kernel_size=(2, 2, 2))

        # Fully connected layers
        self.fc1 = nn.Linear(256 * 2 * 7 * 7, 512)
        self.dropout = nn.Dropout(0.5)
        self.fc2 = nn.Linear(512, num_classes)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        # x shape: (batch, channels, time, height, width)
        x = F.relu(self.bn1(self.conv1(x)))
        x = self.pool1(x)

        x = F.relu(self.bn2(self.conv2(x)))
        x = self.pool2(x)

        x = F.relu(self.bn3(self.conv3(x)))
        x = self.pool3(x)

        # Flatten
        x = x.view(x.size(0), -1)

        x = F.relu(self.fc1(x))
        x = self.dropout(x)
        x = self.fc2(x)

        return x


class ModelLoader:
    """
    Main model loader and manager

    Handles downloading, loading, and managing all detection models.

    Usage:
        loader = ModelLoader(use_gpu=True)
        models = loader.load_all_models()
        spatial_model = models['spatial']
    """

    # Default model configurations
    DEFAULT_MODELS = {
        'spatial_efficientnet': ModelConfig(
            name='spatial_efficientnet',
            repo_id='deepfake-hunter/efficientnet-b4-ff++',
            filename='efficientnet_b4_ffpp.pth',
            model_type='efficientnet',
            input_size=(224, 224),
            version='1.0.0'
        ),
        'temporal_3dcnn': ModelConfig(
            name='temporal_3dcnn',
            repo_id='deepfake-hunter/3dcnn-temporal',
            filename='3dcnn_temporal.pth',
            model_type='3dcnn',
            input_size=(16, 112, 112),  # (time, height, width)
            version='1.0.0'
        ),
    }

    def __init__(self,
                 use_gpu: bool = True,
                 cache_dir: Optional[Path] = None,
                 download_if_missing: bool = True):
        """
        Initialize model loader

        Args:
            use_gpu: Use GPU if available
            cache_dir: Directory for model cache
            download_if_missing: Auto-download models if not cached
        """
        self.device = "cuda" if use_gpu and torch.cuda.is_available() else "cpu"
        self.cache = ModelCache(cache_dir)
        self.download_if_missing = download_if_missing

        logger.info(f"ModelLoader initialized on {self.device}")

    def download_model(self, config: ModelConfig) -> Path:
        """
        Download model from Hugging Face Hub

        Args:
            config: ModelConfig with download information

        Returns:
            Path to downloaded model file
        """
        try:
            logger.info(f"Downloading {config.name} from {config.repo_id}")

            # For now, we'll create placeholder models since we don't have real HF repos
            # In production, this would actually download from HF
            model_path = self.cache.get_cache_path(config.name)

            if not model_path.exists():
                logger.warning(f"Model {config.name} not available on HF Hub (placeholder)")
                # Create a randomly initialized model as placeholder
                if config.model_type == 'efficientnet':
                    model = EfficientNetDetector(pretrained=True)
                elif config.model_type == '3dcnn':
                    model = CNN3DTemporalDetector()
                else:
                    raise ValueError(f"Unknown model type: {config.model_type}")

                # Save placeholder
                torch.save(model.state_dict(), model_path)
                logger.info(f"Created placeholder model: {model_path}")

            return model_path

        except Exception as e:
            logger.error(f"Failed to download model {config.name}: {e}")
            raise

    def load_model(self,
                   model_name: str,
                   config: Optional[ModelConfig] = None) -> nn.Module:
        """
        Load a specific model

        Args:
            model_name: Name of the model to load
            config: Optional custom ModelConfig

        Returns:
            Loaded PyTorch model
        """
        # Check cache first
        cached_model = self.cache.load_from_cache(model_name, self.device)
        if cached_model is not None:
            cached_model.eval()
            return cached_model

        # Get config
        if config is None:
            config = self.DEFAULT_MODELS.get(model_name)
            if config is None:
                raise ValueError(f"Unknown model: {model_name}")

        # Download if needed
        if self.download_if_missing:
            model_path = self.download_model(config)
        else:
            model_path = self.cache.get_cache_path(model_name)
            if not model_path.exists():
                raise FileNotFoundError(f"Model not cached: {model_name}")

        # Create model architecture
        if config.model_type == 'efficientnet':
            model = EfficientNetDetector(pretrained=False)
        elif config.model_type == '3dcnn':
            model = CNN3DTemporalDetector()
        else:
            raise ValueError(f"Unknown model type: {config.model_type}")

        # Load weights
        try:
            state_dict = torch.load(model_path, map_location=self.device)
            model.load_state_dict(state_dict)
            logger.info(f"Loaded {model_name} from {model_path}")
        except Exception as e:
            logger.warning(f"Failed to load state dict: {e}, using initialized model")

        # Move to device
        model = model.to(self.device)
        model.eval()

        # Cache in memory
        self.cache.save_to_cache(model_name, model)

        return model

    def load_all_models(self) -> Dict[str, nn.Module]:
        """
        Load all default models

        Returns:
            Dictionary mapping model names to loaded models
        """
        models = {}

        for model_name in self.DEFAULT_MODELS:
            try:
                models[model_name] = self.load_model(model_name)
            except Exception as e:
                logger.error(f"Failed to load {model_name}: {e}")

        return models

    def verify_model(self, model_name: str) -> bool:
        """
        Verify model integrity using checksum

        Args:
            model_name: Name of model to verify

        Returns:
            True if model passes verification
        """
        model_path = self.cache.get_cache_path(model_name)

        if not model_path.exists():
            return False

        # Compute checksum
        sha256 = hashlib.sha256()
        with open(model_path, 'rb') as f:
            for chunk in iter(lambda: f.read(4096), b''):
                sha256.update(chunk)

        checksum = sha256.hexdigest()

        # Compare with expected (if available)
        config = self.DEFAULT_MODELS.get(model_name)
        if config and config.checksum:
            if checksum != config.checksum:
                logger.warning(f"Checksum mismatch for {model_name}")
                return False

        return True

    def get_model_info(self) -> Dict[str, Any]:
        """
        Get information about available models

        Returns:
            Dictionary with model information
        """
        info = {
            'device': self.device,
            'cache_dir': str(self.cache.cache_dir),
            'models': {}
        }

        for model_name, config in self.DEFAULT_MODELS.items():
            is_cached = self.cache.is_cached(model_name)
            is_verified = self.verify_model(model_name) if is_cached else False

            info['models'][model_name] = {
                'version': config.version,
                'type': config.model_type,
                'cached': is_cached,
                'verified': is_verified,
                'input_size': config.input_size
            }

        return info


class EnsemblePredictor:
    """
    Ensemble multiple models for more robust predictions

    Uses voting or averaging to combine predictions from multiple models
    """

    def __init__(self, models: Dict[str, nn.Module], device: str = "cuda"):
        self.models = models
        self.device = device

        # Set all models to eval mode
        for model in self.models.values():
            model.eval()

        logger.info(f"EnsemblePredictor initialized with {len(models)} models")

    def predict(self,
               x: torch.Tensor,
               method: str = "average") -> torch.Tensor:
        """
        Make ensemble prediction

        Args:
            x: Input tensor
            method: "average" or "voting"

        Returns:
            Ensemble prediction probabilities
        """
        predictions = []

        with torch.no_grad():
            for model_name, model in self.models.items():
                try:
                    pred = model.predict_proba(x) if hasattr(model, 'predict_proba') else torch.softmax(model(x), dim=1)
                    predictions.append(pred)
                except Exception as e:
                    logger.warning(f"Model {model_name} prediction failed: {e}")

        if not predictions:
            raise RuntimeError("All models failed to predict")

        # Combine predictions
        if method == "average":
            # Average probabilities
            ensemble_pred = torch.mean(torch.stack(predictions), dim=0)
        elif method == "voting":
            # Majority voting
            votes = torch.stack([torch.argmax(p, dim=1) for p in predictions])
            ensemble_pred = torch.mode(votes, dim=0).values
        else:
            raise ValueError(f"Unknown ensemble method: {method}")

        return ensemble_pred


# CLI for downloading models
if __name__ == "__main__":
    import argparse

    parser = argparse.ArgumentParser(description="Deepfake Hunter Model Loader")
    parser.add_argument("--download-all", action="store_true",
                       help="Download all models")
    parser.add_argument("--list", action="store_true",
                       help="List available models")
    parser.add_argument("--verify", action="store_true",
                       help="Verify all cached models")
    parser.add_argument("--clear-cache", action="store_true",
                       help="Clear model cache")
    parser.add_argument("--gpu", action="store_true",
                       help="Use GPU if available")

    args = parser.parse_args()

    loader = ModelLoader(use_gpu=args.gpu)

    if args.list:
        info = loader.get_model_info()
        print("\n=== Model Information ===")
        print(f"Device: {info['device']}")
        print(f"Cache Directory: {info['cache_dir']}\n")

        for model_name, model_info in info['models'].items():
            print(f"{model_name}:")
            print(f"  Version: {model_info['version']}")
            print(f"  Type: {model_info['type']}")
            print(f"  Cached: {model_info['cached']}")
            print(f"  Verified: {model_info['verified']}")
            print()

    if args.download_all:
        print("\n=== Downloading All Models ===")
        models = loader.load_all_models()
        print(f"\nSuccessfully loaded {len(models)} models")

    if args.verify:
        print("\n=== Verifying Models ===")
        for model_name in ModelLoader.DEFAULT_MODELS:
            verified = loader.verify_model(model_name)
            status = "✓" if verified else "✗"
            print(f"{status} {model_name}")

    if args.clear_cache:
        print("\n=== Clearing Cache ===")
        loader.cache.clear_cache()
        print("Cache cleared")

    if not any([args.list, args.download_all, args.verify, args.clear_cache]):
        parser.print_help()