Update controlnet_module.py

controlnet_module.py

@@ -998,8 +998,26 @@ class ControlNetProcessor:
                     print(f"   🔍 Analysiere Maske {i+1} auf Crop-Größe")
         
                     # 1. FLÄCHENBASIERTE BEWERTUNG (40%)
-                    area_ratio = mask_area_pixels / bbox_area
+                    area_ratio = mask_area_pixels / bbox_area if bbox_area > 0 else 0
                     print(f"     📐 Flächen-Ratio: {area_ratio:.3f} ({mask_area_pixels:,} / {bbox_area:,} Pixel)")
+
+                    
+                    optimal_min, optimal_max = 0.8, 1.2
+                    if optimal_min <= area_ratio <= optimal_max:
+                        area_score = 1.0
+                    elif area_ratio < optimal_min:
+                        area_score = area_ratio / optimal_min
+                    else:
+                        area_score = optimal_max / area_ratio  # Starke Bestrafung für zu groß
+
+
+                    # 2. BBOX-ÜBERLAPPUNG (KORRIGIERT!)
+                    bbox_mask = np.zeros((image.height, image.width), dtype=np.uint8)
+                    bbox_mask[bbox_y1:bbox_y2, bbox_x1:bbox_x2] = 1
+                    overlap = np.sum(mask_binary & bbox_mask)
+                    bbox_overlap_ratio = overlap / bbox_area if bbox_area > 0 else 0  # WICHTIG: durch bbox_area!
+        
+                    
         
                     # Optimale Kopfgröße: 80-120% der BBox
                     if area_ratio < 0.6:
@@ -1014,6 +1032,7 @@ class ControlNetProcessor:
                     else:
                         # Sanfte Abweichung
                         area_score = 1.0 - abs(area_ratio - 1.0) * 0.5
+                        
         
                     # 2. KOMPAKTHEIT/SOLIDITÄT (30%)
                     labeled_mask = measure.label(mask_binary)