Confidence + Bucket

It crops to the highest confidence score instead of largest box.

The bucket or receipts will now be stored in their correct folders corresponding to their category

app.py

@@ -349,28 +349,62 @@ def parse_total(value: str | float | int | None) -> float:
     return float(m[-1]) if m else 0.0
 
 
+# def detect_and_crop_largest_receipt(img_array: np.ndarray):
+#     """Detect the largest bounding box via YOLO and return crop and bbox.
+
+#     Returns (cropped_array, bbox_dict)
+#     bbox_dict = {"x1": int, "y1": int, "x2": int, "y2": int}
+#     """
+#     results = model(img_array)
+#     # Default fallback to full image
+#     cropped = img_array
+#     bbox = None
+#     for r in results:
+#         if len(r.boxes.xyxy) == 0:
+#             continue
+#         boxes = r.boxes.xyxy
+#         i = max(range(len(boxes)), key=lambda j: (boxes[j][2]-boxes[j][0]) * (boxes[j][3]-boxes[j][1]))
+#         x1, y1, x2, y2 = map(int, boxes[i].tolist())
+#         cropped = img_array[y1:y2, x1:x2]
+#         bbox = {"x1": x1, "y1": y1, "x2": x2, "y2": y2}
+#         break
+#     return cropped, bbox
+
 def detect_and_crop_largest_receipt(img_array: np.ndarray):
-    """Detect the largest bounding box via YOLO and return crop and bbox.
+    """Detect receipt via YOLO and return the crop of the highest-confidence box.
 
     Returns (cropped_array, bbox_dict)
-    bbox_dict = {"x1": int, "y1": int, "x2": int, "y2": int}
+    bbox_dict = {"x1": int, "y1": int, "x2": int, "y2": int, "conf": float}
     """
     results = model(img_array)
-    # Default fallback to full image
     cropped = img_array
     bbox = None
+
     for r in results:
         if len(r.boxes.xyxy) == 0:
             continue
-        boxes = r.boxes.xyxy
-        i = max(range(len(boxes)), key=lambda j: (boxes[j][2]-boxes[j][0]) * (boxes[j][3]-boxes[j][1]))
+
+        boxes = r.boxes.xyxy.cpu().numpy()
+        scores = r.boxes.conf.cpu().numpy()
+
+        # Pick index of highest confidence
+        i = int(scores.argmax())
+
         x1, y1, x2, y2 = map(int, boxes[i].tolist())
         cropped = img_array[y1:y2, x1:x2]
-        bbox = {"x1": x1, "y1": y1, "x2": x2, "y2": y2}
-        break
+        bbox = {
+            "x1": x1,
+            "y1": y1,
+            "x2": x2,
+            "y2": y2,
+            "conf": float(scores[i]),
+        }
+        break  # only use first image in batch
+
     return cropped, bbox
 
 
+
 def encode_image_to_data_url(img_array: np.ndarray, format: str = "JPEG") -> str:
     """Encode an RGB image array to a data URL suitable for GPT-4o image input."""
     pil_img = Image.fromarray(img_array)
@@ -540,11 +574,27 @@ async def process_receipt_vision(
     content_str = getattr(result, "content", str(result))
     extracted_obj = parse_extracted_json(content_str)
 
-    # Upload CROPPED image to Storage (JPEG)
     cropped_bytes = encode_image_to_jpeg_bytes(cropped)
-    storage_path = f"{user_id}/{int(time.time()*1000)}_cropped.jpg"
+
+    # Build storage path with category slug
+    category = extracted_obj.get("category")
+    def category_to_slug(category: str | None) -> str:
+        if not category:
+            return "uncategorized"
+        return (
+            category.strip().lower()
+            .replace("&", "and")
+            .replace("/", "_")
+            .replace(" ", "_")
+            .replace("-", "_")
+        )
+
+    category_slug = category_to_slug(category)
+    storage_path = f"{user_id}/{category_slug}/{int(time.time()*1000)}_cropped.jpg"
+
     try:
         stored = upload_to_storage(storage_path, cropped_bytes, "image/jpeg")
+
     except Exception as exc:
         from fastapi import HTTPException
         raise HTTPException(status_code=502, detail=f"Storage upload failed: {exc}")