File size: 4,838 Bytes
e55a650
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Easy inference script for Fake Image Detection
Usage: python inference.py --image path/to/image.jpg
"""

import torch
from torchvision import transforms
from PIL import Image
import pickle
import json
import argparse
from huggingface_hub import hf_hub_download
from model import EnhancedFreqVAE, EdgeNormalizingFlow, SemanticDeepSVDD, Ensemble


def load_models(device='cuda'):
    """Load all models from Hugging Face"""
    repo_id = "ash12321/fake-image-detection-ensemble"
    
    print("📥 Downloading models from Hugging Face...")
    
    # Load config
    config_path = hf_hub_download(repo_id=repo_id, filename="config.json")
    with open(config_path, 'r') as f:
        config = json.load(f)
    
    # Load PyTorch models
    print("Loading Frequency VAE...")
    freq_vae = EnhancedFreqVAE()
    vae_path = hf_hub_download(repo_id=repo_id, filename="freq_vae.pth")
    freq_vae.load_state_dict(torch.load(vae_path, map_location=device))
    freq_vae.to(device)
    freq_vae.eval()
    
    print("Loading Edge Flow...")
    edge_flow = EdgeNormalizingFlow()
    flow_path = hf_hub_download(repo_id=repo_id, filename="edge_flow.pth")
    edge_flow.load_state_dict(torch.load(flow_path, map_location=device))
    edge_flow.to(device)
    edge_flow.eval()
    
    print("Loading Semantic SVDD...")
    semantic_svdd = SemanticDeepSVDD()
    svdd_path = hf_hub_download(repo_id=repo_id, filename="semantic_svdd.pth")
    checkpoint = torch.load(svdd_path, map_location=device)
    semantic_svdd.load_state_dict(checkpoint['model'])
    semantic_svdd.center = checkpoint['center']
    semantic_svdd.to(device)
    semantic_svdd.eval()
    
    # Load sklearn models
    print("Loading traditional ML models...")
    texture_path = hf_hub_download(repo_id=repo_id, filename="texture_ocsvm.pkl")
    with open(texture_path, 'rb') as f:
        texture_ocsvm = pickle.load(f)
    
    color_path = hf_hub_download(repo_id=repo_id, filename="color_model.pkl")
    with open(color_path, 'rb') as f:
        color_model = pickle.load(f)
    
    stat_path = hf_hub_download(repo_id=repo_id, filename="stat.pkl")
    with open(stat_path, 'rb') as f:
        stat = pickle.load(f)
    
    iforest_path = hf_hub_download(repo_id=repo_id, filename="iforest.pkl")
    with open(iforest_path, 'rb') as f:
        iforest = pickle.load(f)
    
    lof_path = hf_hub_download(repo_id=repo_id, filename="lof.pkl")
    with open(lof_path, 'rb') as f:
        lof = pickle.load(f)
    
    gmm_path = hf_hub_download(repo_id=repo_id, filename="gmm.pkl")
    with open(gmm_path, 'rb') as f:
        gmm = pickle.load(f)
    
    # Create ensemble
    models_dict = {
        'freq_vae': freq_vae,
        'texture_ocsvm': texture_ocsvm,
        'color_model': color_model,
        'edge_flow': edge_flow,
        'semantic_svdd': semantic_svdd,
        'stat': stat,
        'iforest': iforest,
        'lof': lof,
        'gmm': gmm
    }
    
    ensemble = Ensemble(models_dict)
    ensemble.wts = config['weights']
    ensemble.norms = config['norms']
    ensemble.thresh = config['thresh']
    
    print("✓ All models loaded!\n")
    return ensemble, device


def predict_image(image_path, ensemble, device):
    """Predict if an image is fake"""
    # Load and preprocess image
    img = Image.open(image_path)
    img = img.resize((256, 256), Image.LANCZOS).convert('RGB')
    
    tfm = transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize([0.485,0.456,0.406], [0.229,0.224,0.225])
    ])
    img_tensor = tfm(img)
    
    # Predict
    is_fake, score, individual_scores = ensemble.predict(img_tensor, device)
    
    return {
        'prediction': 'FAKE' if is_fake else 'REAL',
        'confidence': abs(score),
        'anomaly_score': score,
        'individual_scores': individual_scores
    }


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description='Detect fake images')
    parser.add_argument('--image', type=str, required=True, help='Path to image')
    parser.add_argument('--device', type=str, default='cuda', help='Device (cuda/cpu)')
    args = parser.parse_args()
    
    # Check device
    device = args.device if torch.cuda.is_available() else 'cpu'
    print(f"Using device: {device}\n")
    
    # Load models
    ensemble, device = load_models(device)
    
    # Predict
    print(f"Analyzing: {args.image}")
    result = predict_image(args.image, ensemble, device)
    
    print("\n" + "="*50)
    print("RESULT")
    print("="*50)
    print(f"Prediction: {result['prediction']}")
    print(f"Confidence: {result['confidence']:.4f}")
    print(f"Anomaly Score: {result['anomaly_score']:.4f}")
    print(f"\nIndividual Model Scores:")
    for model, score in result['individual_scores'].items():
        print(f"  {model}: {score:.4f}")
    print("="*50)