System information (version)

• OpenCV => 4.5.3 and 4.5.4
• Operating System / Platform => Windows 64 bit
• Compiler => Python 3.8

Detailed description

The output of cv.dnn_TextDetectionModel.detectTextRectangles() is different between OpenCV 4.5.3 and OpenCV 4.5.4.

In OpenCV 4.5.3, the output is tuple type ((x, y), (widht, height), angle). That is rotated rectangle format of OpenCV and it can be interpreted correctly by cv.boxPoints().

In OpenCV 4.5.4, the output is numpy.ndarray type [x, y, width, height, angle]. It can not be interpreted correctly by cv.boxPoints(). It will occur an error.

Is this a bug? I think it should return tuple type ((x, y), (widht, height), angle).

Steps to reproduce

1. install opencv-python

   • OpenCV 4.5.3

   pip install opencv-python==4.5.3.56
   

   • OpenCV 4.5.4

   pip install opencv-python==4.5.4.58
   

2. download model file and test image
   The model file (DB_TD500_resnet50.onnx) for text detection is provided in this tutorial.
   The text image (text_det_test1.png) is provided in test data repository.

3. run test program

   import cv2 as cv
   
   print(cv.__version__)
   
   # read image
   image = cv.imread("text_det_test1.png")
   
   # create text detection model
   weights = "DB_TD500_resnet50.onnx"
   text_detection_model = cv.dnn_TextDetectionModel_DB(weights)
   
   # set post-processing params
   text_detection_model.setBinaryThreshold(0.3)
   text_detection_model.setPolygonThreshold(0.5)
   text_detection_model.setMaxCandidates(200)
   text_detection_model.setUnclipRatio(2.0)
   
   # set input shape and normalization params
   text_detection_model.setInputScale(1.0 / 255.0)
   text_detection_model.setInputSize(736, 736)
   text_detection_model.setInputMean((122.67891434, 116.66876762, 104.00698793))
   
   # detect text
   rotated_rectangles, _ = text_detection_model.detectTextRectangles(image)
   
   # print points from rotated rectangles
   for rotated_rectangle in rotated_rectangles:
       print(type(rotated_rectangle))
       print(rotated_rectangle)
       points = cv.boxPoints(rotated_rectangle) # error in opencv 4.5.4
       print(points)

4. output

   • OpenCV 4.5.3

   4.5.3
   <class 'tuple'>
   ((177.50064086914062, 170.92311096191406), (78.48259735107422, 29.635662078857422), -10.561019897460938)
   [[141.63992 192.68217]
   [136.20822 163.54854]
   [213.36136 149.16405]
   [218.79306 178.29768]]
   <class 'tuple'>
   ((226.04806518554688, 118.0884017944336), (202.09889221191406, 52.929649353027344), -12.52880859375)
   [[133.14594  165.84372 ]
   [121.66388  114.174484]
   [318.9502    70.333084]
   [330.43225  122.00232 ]]
   

   • OpenCV 4.5.4

   4.5.4-dev
   <class 'numpy.ndarray'>
   [177.50064  170.92311   78.4826    29.635662 -10.56102 ]
   Traceback (most recent call last):
   File "c:text_detection.py", line 30, in <module>
       points = cv.boxPoints(rotated_rectangle) # error opencv 4.5.4
   cv2.error: OpenCV(4.5.4-dev) :-1: error: (-5:Bad argument) in function 'boxPoints'
   > Overload resolution failed:
   >  - Can't parse 'box' as RotatedRect. Expected sequence length 3, got 5
   >  - Can't parse 'box' as RotatedRect. Expected sequence length 3, got 5
   

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues,
  forum.opencv.org, Stack Overflow, etc and have not found solution
• I updated to latest OpenCV version and the issue is still there
• There is reproducer code and related data files: videos, images, onnx, etc