Add C++ demos (Updated on 2024-06-03) about opencv_zoo

[ryan1288]

@fengyuentau I'm planning to work on the TrackerVIT C++ demo as well. Should be able to produce a much cleaner solution now that I have a better grasp of it 👍

[DaniAffCH]

Hey there, I'm going to work on Human Segmentation

[fengyuentau]

Hello @DaniAffCH , feel free to submit pull requests!

[kekxv]

handpose_estimation_mediapipe. please
Thank you so much.

[dmatveev]

If we revive DNN backend in G-API, it may make sense to write the C++ demos directly in G-API to combine Video, NN, Pre&Post processing all in the same pipeline effectively. Additional points to OpenCV as an application framework. What do you think?

cc: @vpisarev

[fengyuentau]

It would be good to provide additional demos in G-API to construct a complete end-to-end pipeline. But first we need to have DNN backend in G-API, which is not implemented yet right?

cc @ShiqiYu

[LaurentBerger]

@fengyuentau I can manage MobileNet V1 Are you agree?

[fengyuentau]

@LaurentBerger No problems. Pull requests are welcome.

[LaurentBerger]

About DaSiamRPN there is already an example Model source: opencv/samples/dnn/diasiamrpn_tracker.cpp

[fengyuentau]

About DaSiamRPN there is already an example Model source: opencv/samples/dnn/diasiamrpn_tracker.cpp

Yes, but we are going to replace it. We have two GSoC projects ongoing for training a better but also lighter object trackers.

[LaurentBerger]

Yes, but we are going to replace it. We have two GSoC projects ongoing for training a better but also lighter object trackers.

May be it should be mark as in progress

[LaurentBerger]

handpose_estimation_mediapipe. please
Thank you so much.

It's finished here https://github.com/LaurentBerger/opencv_zoo/tree/handpose and here https://github.com/LaurentBerger/opencv_zoo/tree/
@kekxv

[ryan1288]

Hello @fengyuentau , I'm relatively new to open-sourced contributions but I'd love to start contributing to OpenCV / OpenCV-Zoo. I have experience in C++ and Python, applied to ML/CV/Robotics. Are these demos a good starting point for me? Or are there simpler projects that you recommend I get started with?
Is it safe to assume that the corresponding task from #119 needs to be completed before the demo?

[fengyuentau]

@ryan1288 They can be done in parallel. You are welcome to do whatever you are interested in.

[DaniAffCH]

Can I take Facial Expression Recognition? Is there anyone who is already working on it? @fengyuentau

[fengyuentau]

Can I take Facial Expression Recognition? Is there anyone who is already working on it? @fengyuentau

Sure you can. Feel free to submit PR :)

[ryan1288]

@fengyuentau Put up my first draft PR in for this Issue! PR #232 - C++ Demo - Object Detection (NanoDet)

I'm excited to start working on these projects and more 😄. Before I started working on that PR, I played with all the Python and C++ demos from opencv_zoo. I have several thoughts on possible small improvements:

• Object Detection YOLOX
  • demo.cpp
    • Spelling mistakes on line 285
    • Option to save but has no saving functionality
• Object Detection NanoDet
  • demo.py - spelling mistake on line 151
• Object Tracking VitTrack
  • demo.py
    • assert cv.__version__ > "4.9.0" should be assert cv.__version__ >= "4.9.0"
    • No space between text and upper tracking box border
• Optical Flow Estimation Raft
  • demo.py - assert cv.__version__ > "4.9.0" should be assert cv.__version__ >= "4.9.0"
• Image Classification (MobileNet and PPResNet)
  • Could use better documentation README by adding example classifications & providing a link to the class list
• A bunch of models (such as Object detection - YOLOX)
  • The image cv::Mat from cap.open(image_path) is different from imread(image_path), which causes small differences inthe outputs between the Python and C++ demos. I encountered a similar problem when converting Python to C++ for NanoDet.

From C++ Demo for YOLOX

From Python Demo for YOLOX

• Text Detection PPOCR
  • Errors in C++ demo when running ./build/opencv_zoo_text_detection_ppocr -i=../text.jpg when python3 demo.py --input ../text.jpg -v works
    
• Text Recognition CRNN
  • Errors in C++ demo when running ./build/opencv_zoo_text_recognition_crnn --input ../text.jpg when python3 demo.py --input ../text.jpg -v works
    

I think there were more but these are the ones I remember. If these seem like reasonable problems to fix, I'll go work on them and put up some small PRs!

[fengyuentau]

@ryan1288 Hi, thank you for trying out our demos and the detailed investigation! True indeed there are some problems as you found. It would be really helpful if you would like to create several PRs to resolve them.