Upload app.py

app.py

@@ -17,7 +17,7 @@ import base64
 import hashlib
 import asyncio
 from datetime import datetime
-from typing import Optional, Dict, Any, List
+from typing import Optional, Dict, Any, List, Union
 import logging
 
 import cv2
@@ -39,6 +39,7 @@ try:
     print("✅ ONNX Runtime imported successfully")
 except ImportError as e:
     print(f"❌ ONNX Runtime import failed: {e}")
+    ort = None  # Set to None when import fails
     
     # Try PyTorch as fallback
     try:
@@ -179,7 +180,7 @@ class ModelManager:
                 # Load model dengan available backend
                 session = None
                 
-                if ONNX_AVAILABLE:
+                if ONNX_AVAILABLE and ort is not None:
                     # Load ONNX session dengan CPU optimization
                     providers = ['CPUExecutionProvider']
                     session_options = ort.SessionOptions()
@@ -435,8 +436,8 @@ async def handle_pick_the_challenge(data: dict) -> dict:
         
         # Non-Maximum Suppression
         indices = cv2.dnn.NMSBoxes(
-            np.array(boxes), 
-            np.array(confidences), 
+            boxes,  # Use original list instead of numpy array
+            confidences,  # Use original list instead of numpy array
             model_data['confidence'], 
             model_data['nms']
         )
@@ -452,8 +453,53 @@ async def handle_pick_the_challenge(data: dict) -> dict:
         best_match_box = None
         highest_score = 0
         
-        for i in indices.flatten():
-            if class_ids[i] == target_class_id:
+        # Handle indices properly - cv2.dnn.NMSBoxes can return different types
+        indices_flat: List[int] = []
+        if indices is not None and len(indices) > 0:
+            # Convert to list of integers with proper type handling
+            try:
+                # Check if it's a numpy array
+                if isinstance(indices, np.ndarray):
+                    indices_flat = indices.flatten().tolist()
+                elif hasattr(indices, '__iter__') and not isinstance(indices, (str, bytes)):
+                    # Handle iterable (list, tuple, etc.)
+                    temp_list = []
+                    for idx in indices:
+                        if isinstance(idx, (list, tuple, np.ndarray)):
+                            # Nested iterable - flatten it
+                            try:
+                                if isinstance(idx, np.ndarray):
+                                    temp_list.extend(idx.flatten().tolist())
+                                else:
+                                    temp_list.extend([int(x) for x in idx])
+                            except (TypeError, ValueError):
+                                # Skip invalid nested items
+                                continue
+                        else:
+                            # Single value
+                            try:
+                                temp_list.append(int(idx))
+                            except (TypeError, ValueError):
+                                # Skip invalid items
+                                continue
+                    indices_flat = temp_list
+                else:
+                    # Handle single numeric value
+                    try:
+                        # Check if it's numeric
+                        if isinstance(indices, (int, float)):
+                            indices_flat = [int(indices)]
+                        else:
+                            indices_flat = []
+                    except (TypeError, ValueError):
+                        indices_flat = []
+            except Exception as e:
+                # fallback to empty list if conversion fails
+                logger.warning(f"Failed to process NMS indices: {e}")
+                indices_flat = []
+        
+        for i in indices_flat:
+            if 0 <= i < len(class_ids) and class_ids[i] == target_class_id:
                 current_score = confidences[i]
                 if current_score > highest_score:
                     highest_score = current_score
@@ -522,7 +568,7 @@ async def handle_upright_challenge(data: dict) -> dict:
 
         # Determine input size for this inference (prefer model config, default to 300 for upright 3x2 grid)
         input_size = model_data.get('input_size')
-        if not isinstance(input_size) or not isinstance(input_size, int) or input_size <= 0:
+        if not isinstance(input_size, int) or input_size <= 0:
             input_size = 300
 
         input_tensor = preprocess_image(image_bytes, input_size)