Update app.py

app.py

@@ -14,7 +14,7 @@ app = Flask(__name__)
 app.secret_key = 'your_secret_key'  # Replace with a secure secret key
 
 # Global dictionary to hold job progress and results
-jobs = {}  # jobs[job_id] = {"progress": int, "result": {…}}
+jobs = {}  # jobs[job_id] = {"progress": int, "result": {...}}
 
 #########################################
 # 1. Initialize the Models
@@ -90,6 +90,7 @@ def process_image(job_id, image_path, object_type, multiplier):
         # Compute dynamic thickness based on image size and multiplier.
         thickness = max(2, int(min(img_width, img_height) / 300)) * multiplier
         detection_info = []
+        marker_bbox = None  # To store marker bounding box if detected
 
         if object_type == "box":
             if box_model is None:
@@ -97,8 +98,8 @@ 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 is small to help detect small boxes.
-            scale_factor = 1
+            # Upscale the image if it's small so that small boxes become more prominent.
+            scale_factor = 1  # default: no upscaling
             if img_width < 1000 or img_height < 1000:
                 scale_factor = 2
 
@@ -141,7 +142,7 @@ def process_image(job_id, image_path, object_type, multiplier):
                 except Exception as e:
                     continue
 
-            # Use a higher NMS threshold to prevent merging valid nearby detections.
+            # Increase NMS IoU threshold to 0.7 so that nearby valid boxes are not merged.
             box_detections = custom_nms(processed_preds, iou_threshold=0.7)
             jobs[job_id]['progress'] = 60
 
@@ -149,12 +150,12 @@ def process_image(job_id, image_path, object_type, multiplier):
             # We assume the marker is printed at 5cm x 5cm.
             marker_real_width_cm = 5.0
             try:
-                # Create a fresh copy for ArUco detection to avoid drawn overlays
+                # Use a fresh copy for detection to avoid interference from drawn overlays.
                 aruco_image = image.copy()
                 gray = cv2.cvtColor(aruco_image, cv2.COLOR_BGR2GRAY)
-                # Use DICT_6X6_250 – ensure your printed marker is generated with this dictionary.
+                # Use DICT_6X6_250 (ensure your printed marker is generated with this dictionary)
                 aruco_dict = cv2.aruco.getPredefinedDictionary(cv2.aruco.DICT_6X6_250)
-                # Create and adjust detector parameters for small markers
+                # Create detector parameters and adjust for small markers
                 aruco_params = cv2.aruco.DetectorParameters_create()
                 aruco_params.adaptiveThreshWinSizeMin = 3
                 aruco_params.adaptiveThreshWinSizeMax = 23
@@ -167,6 +168,12 @@ def process_image(job_id, image_path, object_type, multiplier):
                 if ids is not None and len(corners) > 0:
                     marker_corners = corners[0].reshape((4, 2))
                     cv2.aruco.drawDetectedMarkers(image, corners, ids)
+                    # Compute bounding box of the marker
+                    mx1 = int(min(marker_corners[:, 0]))
+                    my1 = int(min(marker_corners[:, 1]))
+                    mx2 = int(max(marker_corners[:, 0]))
+                    my2 = int(max(marker_corners[:, 1]))
+                    marker_bbox = (mx1, my1, mx2, my2)
                     side_lengths = []
                     for i in range(4):
                         pt1 = marker_corners[i]
@@ -182,8 +189,14 @@ def process_image(job_id, image_path, object_type, multiplier):
                 print("DEBUG: Exception in ArUco detection:", e)
                 conversion_factor = None
 
+            # Process each detected box
             for pred in box_detections:
                 x1, y1, x2, y2 = pred["box"]
+                # If a marker bounding box was detected, skip boxes that significantly overlap it.
+                if marker_bbox is not None:
+                    iou_with_marker = compute_iou((x1, y1, x2, y2), marker_bbox)
+                    if iou_with_marker > 0.5:
+                        continue
                 label = pred["class"]
                 confidence = pred["confidence"]
                 cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0), thickness)