Update app.py

app.py

@@ -194,47 +194,54 @@ async def predict_single_dog(image):
     
 #     return dogs
 
-async def detect_multiple_dogs(image, conf_threshold=0.35, iou_threshold=0.55, min_area_ratio=0.01, overlap_threshold=0.8):
+async def detect_multiple_dogs(image, conf_threshold=0.35, iou_threshold=0.55):
     results = model_yolo(image, conf=conf_threshold, iou=iou_threshold)[0]
     dogs = []
     boxes = []
-    image_area = image.width * image.height
     
     for box in results.boxes:
         if box.cls == 16:  # COCO dataset class for dog is 16
             xyxy = box.xyxy[0].tolist()
             confidence = box.conf.item()
-            
-            # 計算檢測框面積
-            box_area = (xyxy[2] - xyxy[0]) * (xyxy[3] - xyxy[1])
-            
-            # 如果檢測框面積小於圖片面積的一定比例，則忽略
-            if box_area / image_area < min_area_ratio:
-                continue
-            
             boxes.append((xyxy, confidence))
     
     if not boxes:
         dogs.append((image, 1.0, [0, 0, image.width, image.height]))
     else:
-        # 使用非最大抑制
         nms_boxes = non_max_suppression(boxes, iou_threshold)
         
-        # 進一步合併重疊嚴重的框
-        merged_boxes = merge_overlapping_boxes(nms_boxes, overlap_threshold)
-        
-        for box, confidence in merged_boxes:
-            x1, y1, x2, y2 = box
-            w, h = x2 - x1, y2 - y1
-            x1 = max(0, x1 - w * 0.05)
-            y1 = max(0, y1 - h * 0.05)
-            x2 = min(image.width, x2 + w * 0.05)
-            y2 = min(image.height, y2 + h * 0.05)
+        for box, confidence in nms_boxes:
+            x1, y1, x2, y2 = [int(coord) for coord in box]
             cropped_image = image.crop((x1, y1, x2, y2))
-            dogs.append((cropped_image, confidence, [x1, y1, x2, y2]))
+            
+            if is_valid_dog(cropped_image):
+                dogs.append((cropped_image, confidence, [x1, y1, x2, y2]))
     
     return dogs
 
+def is_valid_dog(image):
+    # 將PIL Image轉換為numpy陣列
+    img_array = np.array(image)
+    
+    # 1. 簡單的紋理檢測
+    gray = np.mean(img_array, axis=2)
+    texture = np.std(gray)
+    
+    # 2. 顏色分布檢測
+    img_rgb = img_array.reshape(-1, 3)
+    kmeans = KMeans(n_clusters=3, n_init=10)
+    kmeans.fit(img_rgb)
+    colors = kmeans.cluster_centers_
+    color_variety = np.std(colors)
+    
+    # 3. 形狀檢測（簡化版，檢查長寬比）
+    aspect_ratio = image.width / image.height
+    
+    # 根據特徵綜合判斷
+    if texture > 10 and color_variety > 30 and 0.5 < aspect_ratio < 2:
+        return True
+    return False
+
 def merge_overlapping_boxes(boxes, overlap_threshold):
     merged = []
     while boxes:
@@ -369,7 +376,7 @@ async def predict(image):
                 explanations.append(dog_explanation)
                 buttons.extend([f"Dog {i+1}: More about {breed}" for breed in topk_breeds[:3]])
             else:
-                explanations.append("The image is unclear or the breed is not in the dataset. Please upload a clearer image.")
+                explanations.append(f"{i+1} The image is unclear or the breed is not in the dataset. Please upload a clearer image.")
                 
 
         final_explanation = "\n\n".join(explanations)