PaddleOCR+OpenCV detection visuals messed up

[4833R11Y45]

class Detection:
"""Represents a single OCR detection as a RECTANGLE (x_min, y_min, x_max, y_max)"""
text: str
bbox: Tuple[int, int, int, int] # axis-aligned rectangle!
confidence: float
tile_offset: Tuple[int, int]

def get_global_bbox(self) -> Tuple[int, int, int, int]:
    x0, y0, x1, y1 = self.bbox
    tx, ty = self.tile_offset
    return (x0+tx, y0+ty, x1+tx, y1+ty)

def get_global_center(self) -> Tuple[float, float]:
    x0, y0, x1, y1 = self.get_global_bbox()
    return ((x0 + x1) / 2, (y0 + y1) / 2)

def run_paddleocr_on_tile(
ocr_engine: PaddleOCR,
tile: np.ndarray,
tile_offset: Tuple[int, int],
debug: bool = False,
debug_all: bool = False
) -> List[Detection]:
"""
Run PaddleOCR 3.3.2 on a tile. Save all output as (x_min, y_min, x_max, y_max) rectangles.
"""
results = list(ocr_engine.predict(tile))
detections = []
if not results:
if debug: print(" [DEBUG] No results returned from PaddleOCR")
return []
result_obj = results[0]
res_dict = None
if hasattr(result_obj, 'json'):
json_dict = result_obj.json
res_dict = json_dict.get('res', {}) if isinstance(json_dict, dict) else {}
elif hasattr(result_obj, 'res'):
res_dict = result_obj.res
if not (isinstance(res_dict, dict) and 'dt_polys' in res_dict):
if debug: print(" [DEBUG] No dt_polys found")
return []
dt_polys = res_dict.get('dt_polys', [])
rec_texts = res_dict.get('rec_texts', [])
rec_scores = res_dict.get('rec_scores', [])
for i, poly in enumerate(dt_polys):
text = rec_texts[i] if i < len(rec_texts) else ""
conf = rec_scores[i] if i < len(rec_scores) else 1.0
if not text.strip():
continue
# Always use axis-aligned rectangle
points = np.array(poly, dtype=np.float32).reshape((-1, 2))
x_min, y_min = np.min(points, axis=0)
x_max, y_max = np.max(points, axis=0)
bbox = (int(x_min), int(y_min), int(x_max), int(y_max))
detections.append(
Detection(text=text, bbox=bbox, confidence=float(conf), tile_offset=tile_offset)
)
return detections

def visualize_detections(floorplan: np.ndarray,
ceiling_detections: List[Detection],
height_detections: List[Detection],
matches: List[CeilingMatch],
output_path: str):
vis_img = floorplan.copy()
for det in ceiling_detections:
x0, y0, x1, y1 = det.get_global_bbox()
cv2.rectangle(vis_img, (x0, y0), (x1, y1), (0, 255, 0), 2)
cv2.putText(vis_img, det.text, (x0, y0 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2)
for det in height_detections:
x0, y0, x1, y1 = det.get_global_bbox()
cv2.rectangle(vis_img, (x0, y0), (x1, y1), (255, 0, 0), 2)
cv2.putText(vis_img, det.text, (x0, y0 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 0, 0), 2)
for match in matches:
cxy = match.ceiling_detection.get_global_center()
hxy = match.height_detection.get_global_center()
cv2.line(vis_img, (int(cxy[0]), int(cxy[1])), (int(hxy[0]), int(hxy[1])), (0, 255, 255), 2)
cv2.imwrite(output_path, cv2.cvtColor(vis_img, cv2.COLOR_RGB2BGR))
print(f" Saved visualization to {output_path}")

I am using PaddleOCR 3.2.2, I would be really thankful if anyone can help.