Update app.py

app.py

@@ -78,7 +78,7 @@ def custom_nms(preds, iou_threshold=0.3):
 def process_image(job_id, image_path, object_type, multiplier):
     try:
         jobs[job_id]['progress'] = 10
-        # Load original image
+        # Load the original image freshly
         image = cv2.imread(image_path)
         if image is None:
             jobs[job_id]['progress'] = 100
@@ -87,7 +87,7 @@ def process_image(job_id, image_path, object_type, multiplier):
 
         jobs[job_id]['progress'] = 20
         img_height, img_width = image.shape[:2]
-        # Compute dynamic thickness based on image size and multiplier.
+        # Compute a dynamic thickness based on image size and multiplier.
         thickness = max(2, int(min(img_width, img_height) / 300)) * multiplier
         detection_info = []
 
@@ -97,17 +97,16 @@ def process_image(job_id, image_path, object_type, multiplier):
                 jobs[job_id]['result'] = {"error": "Roboflow model not available."}
                 return
 
-            # Upscale the image if it's small so that small boxes become more prominent.
-            scale_factor = 1  # default: no upscaling
+            # Upscale the image if it's small (to help detect small boxes)
+            scale_factor = 1
             if img_width < 1000 or img_height < 1000:
-                scale_factor = 2  # adjust factor as needed
+                scale_factor = 2
 
-            # Run detection on the appropriate image version.
             if scale_factor > 1:
                 upscaled_image = cv2.resize(image, None, fx=scale_factor, fy=scale_factor, interpolation=cv2.INTER_LINEAR)
                 temp_path = "upscaled.jpg"
                 cv2.imwrite(temp_path, upscaled_image)
-                # Lower confidence threshold to 30 and overlap to 20 for better sensitivity
+                # Lower the confidence threshold to 30 and overlap to 20 for better sensitivity
                 results = box_model.predict(temp_path, confidence=30, overlap=20).json()
             else:
                 results = box_model.predict(image_path, confidence=30, overlap=20).json()
@@ -117,7 +116,6 @@ def process_image(job_id, image_path, object_type, multiplier):
             for prediction in predictions:
                 try:
                     if scale_factor > 1:
-                        # Adjust coordinates from the upscaled image back to the original scale.
                         x = prediction["x"] / scale_factor
                         y = prediction["y"] / scale_factor
                         width = prediction["width"] / scale_factor
@@ -144,24 +142,21 @@ def process_image(job_id, image_path, object_type, multiplier):
                 except Exception as e:
                     continue
 
-            # Increase the NMS IoU threshold to 0.7 so that touching boxes are less likely to be suppressed.
+            # Increase NMS IoU threshold to 0.7 to prevent suppression of valid, close boxes.
             box_detections = custom_nms(processed_preds, iou_threshold=0.7)
             jobs[job_id]['progress'] = 60
 
-            # (Optional) Detect ArUco marker for measurement.
-            # Set the real-world width of your marker in centimeters.
-            # For a marker that is 5cm x 5cm, use 5.0.
+            # ArUco Marker Detection for measurement:
+            # Here we assume the marker is printed at 5cm x 5cm.
             marker_real_width_cm = 5.0  
             try:
-                gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
-                # Use a predefined dictionary. You can try DICT_6X6_250 (or DICT_4X4_50 if you prefer).
+                # Use a fresh copy of the image (without drawn overlays) for detection.
+                aruco_image = image.copy()
+                gray = cv2.cvtColor(aruco_image, cv2.COLOR_BGR2GRAY)
+                # Use the DICT_6X6_250 dictionary. (You can change to DICT_4X4_50 if desired.)
                 aruco_dict = cv2.aruco.getPredefinedDictionary(cv2.aruco.DICT_6X6_250)
-                # Alternatively:
-                # aruco_dict = cv2.aruco.getPredefinedDictionary(cv2.aruco.DICT_4X4_50)
-                if hasattr(cv2.aruco, 'DetectorParameters_create'):
-                    aruco_params = cv2.aruco.DetectorParameters_create()
-                else:
-                    aruco_params = cv2.aruco.DetectorParameters()
+                aruco_params = cv2.aruco.DetectorParameters_create()
+                # Optionally tweak parameters, e.g., aruco_params.minMarkerPerimeterRate = 0.03
                 corners, ids, _ = cv2.aruco.detectMarkers(gray, aruco_dict, parameters=aruco_params)
                 if ids is not None and len(corners) > 0:
                     # Use the first detected marker as reference.