@gursimarsingh, I found it's almost impossible to run the sample.

1. If I build OpenCV from source and don't run "sudo make install" (which is not recommended anyway), I must set PYTHONPATH pointing to the directory containing cv2.cpython.*.so. It must be described somewhere, otherwise download_models.py cannot be run.
2. download_models.py does not report information where it's downloading models.
3. download_models.py tries to download some models repeatedly, e.g. resnet50. Please, make sure that all hash sums are proper and properly computed. Run download_models.py twice and make sure that once some model was downloaded, it's not downloaded again.
4. How do I specify where my models are stored? I tried to specify OPENCV_DNN_TEST_DATA_PATH, but it still cannot find resnet or squeezenet.
5. I guess, the samples in 5.x should use the new API to output text. Here is how to do it:

   opencv/modules/imgproc/test/test_drawing.cpp

   Line 761 in a2d2ea6

   TEST(Drawing, ttf_text)

6. Can you double-check that the normalization parameters used for various models are still valid? I was under impression that many image classification models from ONNX model zoo use pretty much the same parameters, mostly defined by the dataset (ImageNet).
7. Could you add MobileNet and EfficientNet into models.yml?
8. Could you add option to just reference some entry in models.yml to specify preprocessing? Say, I cannot make the sample to locate resnet properly. Or I want different kind of Resnet (resnet18 or resnet101). Ok, I downloaded resnet by myself and specify explicit path to it. But I don't want to enter all preprocessing parameters manually. This flag could be called "-p" or "--preproc", e.g.

./example_dnn_classification -m <path_to_my_own_resnet> -p resnet -i lena.jpg

some more comments:

1. 'about' message is not printed, whenever I run classification sample without any parameters or with '-h'.
2. I managed to run example by manually specifying path to the downloaded model and the labels and after copying models.yml to the bin directory. I suggest to use a single environment variable to specify OpenCV source root. From this we can automatically find models.yml, the labels and even downloaded models, if they are properly downloaded.
3. If I don't specify path to the labels explicitly, the error message tells me to download opencv_extra. This is not good. In order to run samples, users should not download opencv_extra. The labels are already inside opencv/samples/data/dnn.
4. the old putText API is still used to print the labels. Please, use the new API where real truetype fonts are used.
5. '--model' parameter is not described.

@vpisarev

1. about is now also printed in case alias is missing
2. I propose changing the download_models.py to download the models at a path appended to OPENCV_SOURCE_ROOT environment variable and use that to find the model. At what point should this environment variable needs to be set for the user?
3. OPENCV_SOURCE_ROOT/samples/data/dnn can be added in the search directory in common.hpp
4. I have added the Fontface and rearranged the arguments to use the new API
5. --model paramter is added by:

keys += genPreprocArguments(modelName, zooFile);
parser = CommandLineParser(argc, argv, keys);

It is described if alias is provided. Example: ./bin/example_dnn_classification resnet -h

1. In description the text still mentions 'OPENCV_DNN_TEST_DATA_PATH', not 'OPENCV_SAMPLES_DATA_PATH'.
2. I don't see any information about 'classes' parameter when I run sample without parameters or with '-h'.
3. When models.yml cannot be located, a proper message is printed without throwing CV_Error(). Good!
4. When OPENCV_SAMPLES_DATA_PATH is set, models.yml seems to be located properly, but the file with labels still cannot be located. And in this case CV_Error() is thrown without any clear diagnostics. Why is that?

@asmorkalov, please, merge it. We need to move forward with sample refactoring