Please review both interface design and code. 🙂

Motivation

Stateful kernel in C++ is really useful to many usecases like tracking.
This PR will enable python G-API users to use stateful kernel and expand the usecases in python G-API.

Feature Design

Here is the example video and the example code of this feature.
The example kernel is like this.

class State:
    def __init__(self, tracker, initial_box):
        self.tracker = tracker
        self.initial_box = initial_box

@cv.gapi.kernel(Tracker)
class TrackerImpl:
    @staticmethod
    def setup(input_image_desc, box):
        tracker = cv.TrackerKCF_create()
        return State(tracker, None)

    @staticmethod
    def run(image, box, state):
        if state.initial_box is None:
            state.tracker.init(image, box)
            state.initial_box = box
            return box
        else:
            ok, bbox = state.tracker.update(image)
            return bbox

Basically it create internal state in the setup method and pass it to the G-API framework by returning it.
run method recieves additional state argument and user can handle state with this argument.
It is quite similar to the C++ stateful kernel API GAPI_OCV_KERNEL_ST and difference is setup method return state to the framework. This is the interface that I'd like to ask you to review 🙂

How to run the demo code?

1. Build OpenCV as usual.
2. Run pip install numpy click
3. Run demo program by the following command.

For video input.

python track_video.py --input ${DEVICE_ID}

For video file

python track_video.py --input ${VIDEO_FILE_PATH}

TODO

• Fix interface design
• Add example PR following this PR.
• Add unit test for this feature.

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the original bug report and related work https://github.com/opencv/opencv/pull/17350/files
• [] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
  -> No need for performance
• The feature is well documented and sample code can be built with the project CMake

• I will work on example code later.