File size: 6,982 Bytes
60efa5a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import cv2
import torch
import numpy as np
from PIL import Image
from typing import List, Dict
from collections import Counter
from models.load_models import model_loader

class VideoAnalyzer:
    """Simple, reliable video analyzer for hackathon demo"""
    
    def __init__(self):
        self.model, self.processor = model_loader.load_video_model()
        self.device = model_loader.get_device()
        
        self.face_cascade = cv2.CascadeClassifier(
            cv2.data.haarcascades + 'haarcascade_frontalface_default.xml'
        )
    
    def detect_face(self, frame: np.ndarray) -> Dict:
        """Detect face in frame"""
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        faces = self.face_cascade.detectMultiScale(gray, 1.3, 5)
        
        if len(faces) > 0:
            x, y, w, h = max(faces, key=lambda f: f[2] * f[3])
            face_crop = frame[y:y+h, x:x+w]
            
            return {
                'detected': True,
                'bbox': {'x': int(x), 'y': int(y), 'w': int(w), 'h': int(h)},
                'face_crop': face_crop
            }
        
        return {'detected': False, 'bbox': None, 'face_crop': None}
    
    def predict_deepfake(self, frame: np.ndarray) -> Dict:
        """Predict if frame is deepfake"""
        frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        pil_img = Image.fromarray(frame_rgb)
        
        inputs = self.processor(images=pil_img, return_tensors="pt")
        
        if self.device == "cuda":
            inputs = {k: v.cuda() for k, v in inputs.items()}
        
        with torch.no_grad():
            outputs = self.model(**inputs)
            logits = outputs.logits
            probs = torch.nn.functional.softmax(logits, dim=-1)
        
        fake_prob = probs[0][1].item() if probs.shape[1] > 1 else probs[0][0].item()
        confidence = max(probs[0]).item()
        
        return {
            'fake_score': round(fake_prob, 3),
            'confidence': round(confidence, 3),
            'label': 'fake' if fake_prob > 0.5 else 'real'
        }
    
    def detect_suspicious_regions(self, face: np.ndarray, fake_score: float) -> List[str]:
        try:
            gray = cv2.cvtColor(face, cv2.COLOR_BGR2GRAY)
            h, w = gray.shape
            
            suspicious_regions = []
            
            regions = {
                'eyes': (int(h*0.25), int(h*0.45), int(w*0.15), int(w*0.85)),
                'nose': (int(h*0.40), int(h*0.65), int(w*0.35), int(w*0.65)),
                'mouth': (int(h*0.60), int(h*0.80), int(w*0.30), int(w*0.70)),
                'forehead': (int(h*0.08), int(h*0.28), int(w*0.25), int(w*0.75)),
                'cheeks': (int(h*0.45), int(h*0.70), int(w*0.15), int(w*0.85)),
                'chin': (int(h*0.75), int(h*0.95), int(w*0.30), int(w*0.70))
            }
            
            for region_name, (y1, y2, x1, x2) in regions.items():
                region = gray[y1:y2, x1:x2]
                
                if region.size == 0:
                    continue
                
                suspicious = False
                
                variance = np.var(region)
                if variance < 200 or variance > 8000:
                    suspicious = True
                
                edges = cv2.Canny(region, 50, 150)
                edge_density = np.sum(edges > 0) / edges.size
                if edge_density < 0.05:
                    suspicious = True
                
                if fake_score > 0.7 and variance < 400:
                    suspicious = True
                
                if suspicious:
                    suspicious_regions.append(region_name)
            
            left_half = gray[:, :w//2]
            right_half = np.fliplr(gray[:, w//2:])
            
            min_width = min(left_half.shape[1], right_half.shape[1])
            left_half = left_half[:, :min_width]
            right_half = right_half[:, :min_width]
            
            symmetry_diff = np.mean(np.abs(left_half.astype(float) - right_half.astype(float)))
            
            if symmetry_diff < 10:
                suspicious_regions.append('unnatural_symmetry')
            
            return suspicious_regions if suspicious_regions else ['none']
        
        except Exception as e:
            print(f"Region detection error: {e}")
            return ['analysis_error']
    
    def analyze_interval(self, interval_data: Dict) -> Dict:
        """Analyze all frames in an interval"""
        frames_data = interval_data['video_data']
        
        if not frames_data:
            return {
                'interval_id': interval_data['interval_id'],
                'interval': interval_data['interval'],
                'fake_score': 0.0,
                'confidence': 0.0,
                'suspicious_regions': [],
                'face_detected': False,
                'frame_results': []
            }
        
        frame_results = []
        total_fake_score = 0
        faces_detected = 0
        all_regions = []
        
        for frame_data in frames_data:
            frame = frame_data['frame']
            timestamp = frame_data['timestamp']
            
            face_info = self.detect_face(frame)
            
            if face_info['detected']:
                faces_detected += 1
                pred = self.predict_deepfake(face_info['face_crop'])
                regions = self.detect_suspicious_regions(face_info['face_crop'], pred['fake_score'])
            else:
                pred = self.predict_deepfake(frame)
                regions = ['no_face_detected']
            
            total_fake_score += pred['fake_score']
            all_regions.extend(regions)
            
            frame_results.append({
                'timestamp': timestamp,
                'fake_score': pred['fake_score'],
                'confidence': pred['confidence'],
                'face_detected': face_info['detected'],
                'regions': regions
            })
        
        avg_fake_score = total_fake_score / len(frames_data)
        
        region_counts = Counter(all_regions)
        threshold = len(frames_data) * 0.5
        
        consistent_regions = [
            region for region, count in region_counts.items()
            if count >= threshold and region not in ['none', 'no_face_detected', 'analysis_error']
        ]
        
        return {
            'interval_id': interval_data['interval_id'],
            'interval': interval_data['interval'],
            'fake_score': round(avg_fake_score, 3),
            'confidence': round(np.mean([f['confidence'] for f in frame_results]), 3),
            'suspicious_regions': consistent_regions if consistent_regions else list(set(all_regions)),
            'face_detected': faces_detected > 0,
            'frame_count': len(frames_data),
            'frames_with_faces': faces_detected,
            'frame_results': frame_results
        }