Update main.py

main.py

@@ -74,11 +74,11 @@ class ContentModerator:
         return {
             'weapon_detection': {
                 'enabled': True,
-                'confidence_threshold': 0.45,  # For guns
-                'knife_confidence': 0.45,  # Lower threshold for knives
-                'fight_confidence': 0.40,  # Lower threshold for fights (behavioral)
-                'model_size': 'yolo11n',
-                'classes': ['knife', 'gun', 'rifle', 'pistol', 'weapon', 'fight'],
+                'confidence_threshold': 0.5,  # For guns
+                'knife_confidence': 0.5,  # Lower threshold for knives
+                'fight_confidence': 0.45,  # Lower threshold for fights (behavioral)
+                'model_size': 'yolo12n',
+                'classes': ['gun', 'knife', 'fight'],
                 'use_enhancement': True,  # Enable image enhancement for knives
                 'multi_pass': True,  # Enable multi-pass detection
                 'boost_knife_detection': True,  # Enable knife confidence boosting
@@ -87,10 +87,10 @@ class ContentModerator:
             },
             'fight_detection': {
                 'enabled': True,
-                'confidence_threshold': 0.40,
+                'confidence_threshold': 0.45,
                 'pose_analysis': True,  # Analyze poses for fighting
                 'motion_analysis': False,  # Motion-based fight detection (for video)
-                'aggression_keywords': ['fight', 'violence', 'aggression', 'combat'],
+                'aggression_keywords': ['fight'],
                 'threat_escalation': True,  # Escalate threat level for fights
                 'multi_person_analysis': True  # Analyze interactions between people
             },
@@ -141,7 +141,7 @@ class ContentModerator:
             print("🔫 Loading weapon and fight detection models...")
 
             # Model 1: Custom YOLO11 for weapons (dao + súng + fight)
-            custom_model_path = "models/best.pt"
+            custom_model_path = "models/best_ft4.pt"
             project_root = os.path.dirname(os.path.abspath(__file__))
             full_model_path = os.path.join(project_root, custom_model_path)
 
@@ -204,7 +204,16 @@ class ContentModerator:
             if self.config['weapon_detection']['use_enhancement']:
                 enhanced_image = self.enhance_knife_detection(image)
                 images_to_process.append((enhanced_image, 1.15, "enhanced"))
-
+            if 'fight_detection' not in self.config:
+                self.config['fight_detection'] = {
+                    'enabled': True,
+                    'confidence_threshold': 0.40,
+                    'pose_analysis': False,
+                    'motion_analysis': False,
+                    'aggression_keywords': ['fight'],
+                    'threat_escalation': True,
+                    'multi_person_analysis': False
+                }
             # Process each image version
             for img, weight_multiplier, img_type in images_to_process:
                 if self.weapon_model_custom:
@@ -564,8 +573,6 @@ class ContentModerator:
 
         return detections
 
-    # ... (rest of the existing methods remain the same) ...
-
     def enhance_knife_detection(self, image):
         """Enhance image specifically for better knife/dao detection"""
         try:
@@ -823,8 +830,6 @@ class ContentModerator:
 
         return intersect_area / union_area if union_area > 0 else 0
 
-    # ... (continue with remaining NSFW detection methods) ...
-
     def detect_nsfw_content(self, image):
         """Enhanced NSFW detection with person detection first"""
         detections = []
@@ -1127,8 +1132,14 @@ class ContentModerator:
         """Clean expired cache entries"""
         try:
             expired_keys = []
-            for key, (_, timestamp) in self.detection_cache.items():
-                if current_time - timestamp > self.cache_ttl:
+            for key, value in self.detection_cache.items():
+                # Check tuple structure
+                if isinstance(value, tuple) and len(value) == 2:
+                    _, timestamp = value
+                    if timestamp is not None and current_time - timestamp > self.cache_ttl:
+                        expired_keys.append(key)
+                else:
+                    # Invalid cache entry, remove it
                     expired_keys.append(key)
 
             for key in expired_keys:
@@ -1281,14 +1292,13 @@ class ContentModerator:
             print(f"❌ Error drawing detections: {e}")
             return image
 
-    def process_video(self, video_path, output_path=None, frame_skip=2):
-        """Process video file with enhanced detection including fights - optimized frame processing"""
+    def process_video(self, video_path, output_path=None):
+        """Process video file with enhanced detection including fights - processes every frame"""
         try:
             cap = cv2.VideoCapture(video_path)
             frame_count = 0
             total_detections = []
             fight_timeline = []  # Track fights over time
-            recent_detections = []  # Track recent detections for adaptive processing
 
             if output_path:
                 fourcc = cv2.VideoWriter_fourcc(*'mp4v')
@@ -1304,22 +1314,7 @@ class ContentModerator:
 
                 frame_count += 1
 
-                # Adaptive frame processing based on recent detections
-                should_process = False
-
-                # Always process if recent threats detected (within last 10 frames)
-                if any(det['frame'] > frame_count - 10 for det in recent_detections[-5:]):
-                    should_process = True
-                # Or process based on reduced skip rate
-                elif frame_count % max(1, frame_skip) == 0:
-                    should_process = True
-
-                if not should_process:
-                    if output_path:
-                        out.write(frame)
-                    continue
-
-                # Process frame
+                # Process every frame
                 result = self.process_image(frame)
                 if result and result['detections']:
                     # Add frame number to each detection for tracking
@@ -1327,7 +1322,6 @@ class ContentModerator:
                         detection['frame'] = frame_count
 
                     total_detections.extend(result['detections'])
-                    recent_detections.append({'frame': frame_count, 'count': len(result['detections'])})
 
                     # Track fight timeline
                     fight_detections = [d for d in result['detections'] if d['type'] == 'fight']
@@ -1340,9 +1334,6 @@ class ContentModerator:
                             'max_aggression': max(f.get('aggression_level', 'low') for f in fight_detections)
                         })
 
-                        # Reduce frame_skip temporarily when fight detected
-                        frame_skip = max(1, frame_skip // 2)
-
                     print(f"⚠️ Frame {frame_count}: {len(result['detections'])} threats detected")
 
                     breakdown = result.get('detection_breakdown', {})
@@ -1354,10 +1345,6 @@ class ContentModerator:
                     elif output_path:
                         out.write(frame)
                 else:
-                    # No detections - can increase frame_skip for efficiency
-                    if len(recent_detections) > 5 and all(det['count'] == 0 for det in recent_detections[-5:]):
-                        frame_skip = min(5, frame_skip + 1)
-
                     if output_path:
                         out.write(frame)
 
@@ -1378,7 +1365,7 @@ class ContentModerator:
                 }
 
             return {
-                'total_frames_processed': frame_count // frame_skip,
+                'total_frames_processed': frame_count,
                 'total_detections': len(total_detections),
                 'detections': total_detections,
                 'fight_timeline': fight_timeline,
@@ -1413,7 +1400,7 @@ class ContentModerator:
 
 
 def main():
-    """Enhanced example usage with knife and fight detection improvements"""
+    """Enhanced example usage with knife and fight detection improvements - processes every frame"""
 
     # Initialize the system
     moderator = ContentModerator()
@@ -1544,7 +1531,7 @@ def main():
         result = moderator.process_image(test_img)
         print("✅ Test pattern processed successfully")
 
-    # Example 2: Enhanced webcam processing with fight detection
+    # Example 2: Enhanced webcam processing with fight detection - processes every frame
     print("\n" + "=" * 60)
     print("📹 ENHANCED WEBCAM PROCESSING WITH FIGHT DETECTION")
     print("=" * 60)
@@ -1577,11 +1564,6 @@ def main():
                 'fight_incidents': 0
             }
 
-            # Adaptive processing variables
-            process_interval = 2  # Start with every 2nd frame
-            last_detection_frame = 0
-            consecutive_safe_frames = 0
-
             while True:
                 ret, frame = cap.read()
                 if not ret:
@@ -1610,81 +1592,52 @@ def main():
                 cv2.putText(frame, model_info, (10, y_offset + 60),
                             cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
 
-                # Adaptive frame processing - process more frequently when threats detected
-                should_process = False
-
-                # Always process if recent threats (within last 30 frames)
-                if frame_count - last_detection_frame <= 30:
-                    should_process = (frame_count % 1 == 0)  # Process every frame
-                    cv2.putText(frame, "HIGH ALERT MODE", (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
-                # Normal processing with reduced interval
-                elif frame_count % process_interval == 0:
-                    should_process = True
-
-                if should_process:
-                    result = moderator.process_image(frame)
-
-                    if result and result['action_required']:
-                        last_detection_frame = frame_count  # Update last detection frame
-                        consecutive_safe_frames = 0
-                        process_interval = 1  # Process every frame when threats detected
-
-                        # Count detections by type
-                        for detection in result['detections']:
-                            if detection['type'] == 'weapon':
-                                detection_stats['weapons'] += 1
-                                if detection['weapon_type'] == 'blade':
-                                    detection_stats['knives'] += 1
-                                elif detection['weapon_type'] == 'firearm':
-                                    detection_stats['guns'] += 1
-                            elif detection['type'] == 'fight':
-                                detection_stats['fights'] += 1
-                                if detection.get('aggression_level') in ['high', 'extreme']:
-                                    detection_stats['fight_incidents'] += 1
-                            elif detection['type'] == 'nsfw':
-                                detection_stats['nsfw'] += 1
-
-                        print(
-                            f"⚠️ Frame {frame_count}: {result['risk_level']} risk - {result['total_threats']} threats!")
-
-                        # Show specific detections with fight info
-                        for detection in result['detections']:
-                            if detection['type'] == 'weapon':
-                                icon = "🔪" if detection['weapon_type'] == 'blade' else "🔫"
-                                method = detection.get('detection_method', 'unknown').split('_')[-1]
-                                print(f"   {icon} {detection['class']} ({detection['confidence']:.3f}) [{method}]")
-                            elif detection['type'] == 'fight':
-                                fight_type = detection.get('fight_type', 'general')
-                                aggression = detection.get('aggression_level', 'unknown')
-                                people = detection.get('people_involved', 0)
-                                method = detection.get('detection_method', 'unknown').split('_')[-1]
-                                print(f"   👊 FIGHT: {fight_type} ({detection['confidence']:.3f}) [{method}]")
-                                print(f"      Aggression: {aggression}, People: {people}")
-
-                        # Use annotated frame
-                        if 'annotated_image' in result:
-                            cv2.imshow('Enhanced Detection System (Weapons + Fights)', result['annotated_image'])
-                        else:
-                            # Add threat counter
-                            breakdown = result.get('detection_breakdown', {})
-                            threat_text = f"THREATS: W:{breakdown.get('weapons', 0)} F:{breakdown.get('fights', 0)} N:{breakdown.get('nsfw', 0)}"
-                            cv2.putText(frame, threat_text, (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
-                            cv2.imshow('Enhanced Detection System (Weapons + Fights)', frame)
+                # Process every frame
+                result = moderator.process_image(frame)
+
+                if result and result['action_required']:
+                    # Count detections by type
+                    for detection in result['detections']:
+                        if detection['type'] == 'weapon':
+                            detection_stats['weapons'] += 1
+                            if detection['weapon_type'] == 'blade':
+                                detection_stats['knives'] += 1
+                            elif detection['weapon_type'] == 'firearm':
+                                detection_stats['guns'] += 1
+                        elif detection['type'] == 'fight':
+                            detection_stats['fights'] += 1
+                            if detection.get('aggression_level') in ['high', 'extreme']:
+                                detection_stats['fight_incidents'] += 1
+                        elif detection['type'] == 'nsfw':
+                            detection_stats['nsfw'] += 1
+
+                    print(f"⚠️ Frame {frame_count}: {result['risk_level']} risk - {result['total_threats']} threats!")
+
+                    # Show specific detections with fight info
+                    for detection in result['detections']:
+                        if detection['type'] == 'weapon':
+                            icon = "🔪" if detection['weapon_type'] == 'blade' else "🔫"
+                            method = detection.get('detection_method', 'unknown').split('_')[-1]
+                            print(f"   {icon} {detection['class']} ({detection['confidence']:.3f}) [{method}]")
+                        elif detection['type'] == 'fight':
+                            fight_type = detection.get('fight_type', 'general')
+                            aggression = detection.get('aggression_level', 'unknown')
+                            people = detection.get('people_involved', 0)
+                            method = detection.get('detection_method', 'unknown').split('_')[-1]
+                            print(f"   👊 FIGHT: {fight_type} ({detection['confidence']:.3f}) [{method}]")
+                            print(f"      Aggression: {aggression}, People: {people}")
+
+                    # Use annotated frame
+                    if 'annotated_image' in result:
+                        cv2.imshow('Enhanced Detection System (Weapons + Fights)', result['annotated_image'])
                     else:
-                        consecutive_safe_frames += 1
-                        # Gradually increase processing interval when safe (up to max 3)
-                        if consecutive_safe_frames > 30:
-                            process_interval = min(3, process_interval + 1)
-                            consecutive_safe_frames = 0
-
-                        cv2.putText(frame, "SAFE", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
-                        cv2.putText(frame, f"Process Interval: {process_interval}", (10, 90),
-                                    cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
+                        # Add threat counter
+                        breakdown = result.get('detection_breakdown', {})
+                        threat_text = f"THREATS: W:{breakdown.get('weapons', 0)} F:{breakdown.get('fights', 0)} N:{breakdown.get('nsfw', 0)}"
+                        cv2.putText(frame, threat_text, (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
                         cv2.imshow('Enhanced Detection System (Weapons + Fights)', frame)
                 else:
                     cv2.putText(frame, "SAFE", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
-                    cv2.putText(frame, f"Process Interval: {process_interval}", (10, 90),
-                                cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
                     cv2.imshow('Enhanced Detection System (Weapons + Fights)', frame)
 
                 # Handle key presses
@@ -1778,3 +1731,4 @@ def main():
     print("   - Dynamic confidence thresholds")
     print("   - Geometric feature analysis")
     print("   - Multi-pass detection strategy")
+    print("💡 Processing mode: EVERY FRAME (no skipping)")