maicg01/daisykit

DaisyKit is an easy AI toolkit with face mask detection, pose detection, background matting, barcode detection, face recognition and more. - with NCNN, OpenCV, Python wrappers

DaisyKit - D.A.I.S.Y: Deploy AI Systems Yourself!

DaisyKit is an easy AI toolkit with face mask detection, pose detection, background matting, barcode detection and more. This open source project includes:

• DaisyKit SDK - C++, the core of models and algorithms in NCNN deep learning framework.
• DaisyKit Python wrapper for easy integration with Python.
• DaisyKit Android - Example app demonstrate how to use Daisykit SDK in Android.

Links:

• Python Package: https://pypi.org/project/daisykit/
• Documentation: https://daisykit.nrl.ai/docs

Demo Video: https://www.youtube.com/watch?v=zKP8sgGoFMc.

1. Environment Setup

Ubuntu

Install packages from Terminal

sudo apt install -y build-essential libopencv-dev
sudo apt install -y libvulkan-dev vulkan-utils
sudo apt install -y mesa-vulkan-drivers # For Intel GPU support

Windows

For Windows, Visual Studio 2019 + Git Bash is recommended.

• Download and extract OpenCV from the official website, and add OpenCV_DIR to path.
• Download precompiled NCNN.

2. Build and run C++ examples

Clone the source code:

git clone https://github.com/nrl-ai/daisykit.git --recursive
cd daisykit

Ubuntu

Build Daisykit:

mkdir build
cd build
cmake .. -Dncnn_FIND_PATH="<path to ncnn lib>"
make

Run face detection example:

./bin/demo_face_detector_graph

If you dont specify ncnn_FIND_PATH, NCNN will be built from scratch.

Windows

Build Daisykit:

mkdir build
cd build
cmake -G "Visual Studio 16 2019" -Dncnn_FIND_PATH="<path to ncnn lib>" ..
cmake --build . --config Release

Run face detection example:

./bin/Release/demo_face_detector_graph

3. C++ Coding convention

Read coding convention and contribution guidelines here.

4. Known issues and problems

• Slow model inference - Low FPS

This issue can happen on development build. Add -DCMAKE_BUILD_TYPE=Debug to cmake command and build again. The FPS can be much better.

5. References

This toolkit is developed on top of other source code. Including

• Toolchains setup from ncnn.
• QR Scanner from ZXing-CPP.
• JSON support from nlohmann/json.
• Pretrained AI models from different sources: https://daisykit.nrl.ai/docs/models.