@UnaNancyOwen Thank you for contribution. Can you add simple tests to verify Python bindings functionality for exported classes are fixed?

@VadimLevin Sorry, That could be difficult for me. I don't know how python / dnn-module testing in OpenCV.

@VadimLevin Sorry, That could be difficult for me. I don't know how python / dnn-module testing in OpenCV.

You can add a test to modules/dnn/misc/python/test/test_dnn.py as test_high_level_api(self) method of class dnn_test. To reproduce the issue before fix I think you can use models from opencv_extra repository. Before applied fix test should fail and after applied fix test should pass.

@VadimLevin I have written a test. Please check following.
And, I have one question. Do I need to upload model (.onnx) and reference (.npy) to opencv/opencv_extra?

def test_textdetection_model_db(self):
    img_path = self.find_dnn_file("dnn/text_det_test1.png")
    weights = self.find_dnn_file("onnx/models/DB_TD500_resnet50.onnx", required=False)
    ref = np.load(self.find_dnn_file("dnn/DB_TD500_resnet50_prob.npy"))
    if weights is None:
        raise unittest.SkipTest("Missing DNN test files (onnx/models/DB_TD500_resnet50.onnx). Verify OPENCV_DNN_TEST_DATA_PATH configuration parameter.")

    frame = cv.imread(img_path)
    model = cv.dnn_TextDetectionModel_DB(weights)
    model.setInputScale(1.0 / 255.0)
    model.setInputSize(736, 736)
    model.setInputMean(122.67891434, 116.66876762, 104.00698793)

    out = model.predict(frame)
    normAssert(self, out, ref)

@VadimLevin I have written a test. Please check following.
And, I have one question. Do I need to upload model (.onnx) and reference (.npy) to opencv/opencv_extra?

That's my suggestion to use model from opencv_extra, because the issue is related to the generated bindings and can be reproduced with any model. To avoid generating reference npy file you can use corresponding low level API cv::dnn::Net and compare results - they should be close to each other.

That's my suggestion to use model from opencv_extra, because the issue is related to the generated bindings and can be reproduced with any model.

(I don't understand how it works, but) is it okay if the model is described in opencv_extra/testdata/dnn/download_models.py?
DB_TD500_resnet50.onnx is described in here.
https://github.com/opencv/opencv_extra/blob/4.5.1/testdata/dnn/download_models.py#L872-L876

To avoid generating reference npy file you can use corresponding low level API cv::dnn::Net and compare results - they should be close to each other.

That's a good idea. I took that idea in test code.

def test_textdetection_model_db(self):
    img_path = self.find_dnn_file("dnn/text_det_test1.png")
    weights = self.find_dnn_file("dnn/onnx/models/DB_TD500_resnet50.onnx", required=False)
    if weights is None:
        raise unittest.SkipTest("Missing DNN test files (onnx/models/DB_TD500_resnet50.onnx). Verify OPENCV_DNN_TEST_DATA_PATH configuration parameter.")

    frame = cv.imread(img_path)
    scale = 1.0 / 255.0
    size = (736, 736)
    mean = (122.67891434, 116.66876762, 104.00698793)

    model = cv.dnn_TextDetectionModel_DB(weights)
    model.setInputParams(scale, size, mean)
    out = model.predict(frame)

    net = cv.dnn.readNet(weights)
    blob = cv.dnn.blobFromImage(frame, scale, size, mean)
    net.setInput(blob)
    ref = net.forward()

    normAssert(self, out, ref)

To avoid generating reference npy file you can use corresponding low level API cv::dnn::Net and compare results

There is strong recommendation to avoid accuracy checking in Python (and others) bindings tests.
Check only parameters passing and types/shapes of returned results (check binding shim code only).

@alalek I will fix test to check type and shape of output.