Skip to content
/ ncnnRay Public

ncnnRay++ is a CMake based integration of raylib and the very popular Tencent ncnn Deep Learning library. ncnn is written in C++ designed (but not only) for edge computing devices. The project depends on the Vulkan SDK (Vulakn is Khronos' API for Graphics and Compute on GPUs).

1 star 6 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

stjordanis/ncnnRay

BranchesTags

Repository files navigation

ncnnRay++: A CMake / WASM integration of rayib UI and the Tencent ncnn C++ AI platform.

About • Credits • Installation • Examples • Author • License

About

Live demo here: https://quantscientist.github.io/ncnnRay/

ncnnRay++ is a CMake based integration of raylib (https://www.raylib.com/examples.html) and the very popular ncnn (https://github.com/Tencent/ncnn) Deep Learning library. ncnn is written in C++ designed (but not only) for edge computing devices. The project depends on the Vulkan SDK (Vulakn is Khronos' API for Graphics and Compute on GPUs).

There is no OpenCV dependency for reading and writing images / videos or anything else. Everything is self-contained using stb_image . The library includes conversion utils between ray Image, ncnn::Mat and PNG lib, all of which are interchangeable. For more details, refer to:

Why??++ The goal is to integrate both libraries into HTML using Webassembly. ray is already natively supported on WASM. This would open the door to countless opportunities in creativity.

Real-time face detection on a GPU (Vulkan):

Available AI Models (Tencent ncnn in C++)

Most of the models were originally trained in PyTorch, then converted to onnx and then to ncnn.

Image Model name Credit C++ implementation
 NCNN_DEVICE
 Paper
LFFD Credit ncnn C++ and LFFD: A Light and Fast Face Detector for Edge Devices CPU + GPU (Vulkan) LFFD: A Light and Fast Face Detector for Edge Devices
NeuralStyle Credit ncnn C++ CPU / Vulkan GPU A Neural Algorithm of Artistic Style
RetinaFace Credit ncnn C++ CPU + GPU (Vulkan) RetinaFace
More comming soon

GPU or CPU?

If you want to test on a CPU you will have to edit the CMakeLists file. ncnn uses the Vulkan SDK for all its GPU operations.

GPU (Vulkan) mode:

set(NCNN_DEVICE "VULKAN")

CPU mode:

set(NCNN_DEVICE "cpu")

Note: Tested only on GPU with CUDA 10.2 and a CPU on a Windows 10 and Ubuntu 18 machines.

Vulkan

Vulkan is Khronos' API for High-efficiency Graphics and Compute on GPUs.

In analogy to PyTorch which uses CUDA behind the scenes to allocate tensors on a GPU, ncnn uses the Vulkan SDK and shaders to do the same. To use Vulkan backend during CMake build, install Vulkan header files, a vulkan driver loader, GLSL to SPIR-V compiler and vulkaninfo tool. Preferably from your distribution repositories. Alternatively download and install full Vulkan SDK (about 200MB in size; it contains all header files, documentation and prebuilt loader, as well some extra tools and source code of everything) from https://vulkan.lunarg.com/sdk/home

To use Vulkan after building ncnn you will also need to have Vulkan driver for your GPU. For AMD and Intel GPUs these can be found in Mesa graphics driver, which usually is installed by default on all distros. For Nvidia GPUs the propietary Nvidia driver must be downloaded and installed After installing Vulkan driver, confirm Vulkan libraries and driver are working, by using vulkaninfo it should list GPU device type. For instance:

VkPhysicalDeviceDriverProperties:
---------------------------------
driverID           = DRIVER_ID_NVIDIA_PROPRIETARY
driverName         = NVIDIA
driverInfo         = 452.28
conformanceVersion = 1.2.3.1

RayLib

rayLib is an amazing library which has been widely adopted by the gaming community. Read more about the raylib game framework here: https://www.raylib.com/ Or look up projects using it here: https://www.google.com/search?q=raylib+site:github.com

Tencent ncnn

Written in C++, ncnn has a community that is growing by the day. It compiles on almost every platform and is mainly targeted at egde devices. https://github.com/Tencent/ncnn

(Preview)

A simple example

The folowing example create a ray window, allocates a ncnn::Mat on the GPU and draws the value into a ray window.

int main(int argc, char** argv) {
    Image image = LoadImage("faces.png");   // Loaded in CPU memory (RAM)
    ncnn::Mat inmat = rayImageToNcnn(image);
    std::cout << "Total:" << inmat.total() << std::endl;
    std::cout << "D:" << tensorDIMS(inmat) << std::endl;;
    Image saveImage = ncnnToRayImage(inmat);
    ExportImage(saveImage, "faces-ncnn-rgb.png");

    Image imageRGBA = LoadImage("manga.png");   // Loaded in CPU memory (RAM)
    ncnn::Mat inmatimageRGBA = rayImageToNcnn(imageRGBA);
    std::cout << "Total:" << inmatimageRGBA.total() << std::endl;
    std::cout << "D:" << tensorDIMS(inmatimageRGBA) << std::endl;;
    Image saveImageimageRGBA = ncnnToRayImage(inmatimageRGBA);
    ExportImage(saveImageimageRGBA, "manga-ncnn-rgba.png");    

    return 0;

Credits + Third party libraries I used

Progress / features

🔰 ncnnRay++ CMake
ncnn:Mat CPU tensor to stb Image ✔️
ncnn:Mat GPU tensors to stb image ✔️
stb image to CPU ncnn:Mat ✔️
Save ncnn:Mat as stb PNG image on disk ✔️

Examples

A Simple example, mainly for testing the integration. Read an Image, allocate an ncnn:Mat tensor on the CPU.

LFFD: A Light and Fast Face Detector for Edge Devices

@inproceedings{LFFD,
title={LFFD: A Light and Fast Face Detector for Edge Devices},
author={He, Yonghao and Xu, Dezhong and Wu, Lifang and Jian, Meng and Xiang, Shiming and Pan, Chunhong},
booktitle={arXiv:1904.10633},
year={2019}
}

Requirements:

  • Tested only on Windows 10 and Microsoft Visual C++ 2019 16.4
  • Vulkan SDK
  • ncnn (downloaded automatically)
  • Tested on 64 bit only device.
  • CMake 3.18
  • rayLib GUI (downloaded automatically).
  • Vulkan SDK
  • Emscripten emsdk if you want to compile to WASM

Building using CMake

Start by:

git clone --recurssive https://github.com/QuantScientist/ncnnRay.git

ncnn

CMake should take care of everything for you!

cd to 3rdparty\ncnn\
git submodule update --init
call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build\vcvars64.bat"
cmake.exe -DCMAKE_BUILD_TYPE=Release -G "CodeBlocks - NMake Makefiles" . 
-B bin64 -DNCNN_VULKAN=ON -DNCNN_SYSTEM_GLSLANG=ON -DNCNN_BUILD_EXAMPLES=ON -DNCNN_BENCHMARK=ON 
-DNCNN_BUILD_TOOLS=ON  && cmake.exe --build bin64

CLion is strongly recommended for the build.

Then cd to the root folder of the project and run again:

cd to the root folder
call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build\vcvars64.bat"
cmake.exe -DCMAKE_BUILD_TYPE=Release -G "CodeBlocks - NMake Makefiles" . 
-B bin64 -DNCNN_VULKAN=ON -DNCNN_SYSTEM_GLSLANG=ON -DNCNN_BUILD_EXAMPLES=ON -DNCNN_BENCHMARK=ON 
-DNCNN_BUILD_TOOLS=ON  && cmake.exe --build bin64

Downloading and installing steps rayLib:

The CMake file will download this automatically for you.

Compiling on Windows

standalone.bat

Compiling on Linux

Please use the provided docker that includes everything you need including Emscripten / emsdk.

git submodule update --init
cmake -DCMAKE_BUILD_TYPE=Release -G "CodeBlocks - Unix Makefiles" . -B bin64 -DCMAKE_INSTALL_PREFIX=bin64 -DNCNN_VULKAN=OFF -DNCNN_SYSTEM_GLSLANG=OFF -DNCNN_BUILD_EXAMPLES=ON -DNCNN_BENCHMARK=ON -DNCNN_BUILD_TOOLS=ON  && cmake --build bin64
[ 91%] Building CXX object CMakeFiles/ncnnRay_studio_UI.dir/src/models/LFFD.cpp.o
In file included from /src/ncnnRay/ncnnRay/src/models/LFFD.cpp:1:0:
/src/ncnnRay/ncnnRay/src/models/LFFD.h:6:1: warning: 'typedef' was ignored in this declaration
 typedef struct FaceInfo {
 ^~~~~~~
[ 91%] Building CXX object CMakeFiles/ncnnRay_studio_UI.dir/src/models/FaceDetector.cpp.o
[ 91%] Building CXX object CMakeFiles/ncnnRay_studio_UI.dir/src/models/neural.cpp.o
[ 92%] Linking CXX executable ncnnRay_studio_UI
[ 92%] Built target ncnnRay_studio_UI
Scanning dependencies of target ncnnRay_retinaface
[ 92%] Building CXX object CMakeFiles/ncnnRay_retinaface.dir/src/ncnnRay_retinaface.cpp.o
[ 93%] Building CXX object CMakeFiles/ncnnRay_retinaface.dir/src/models/FaceDetector.cpp.o
[ 93%] Linking CXX executable ncnnRay_retinaface
[ 93%] Built target ncnnRay_retinaface
Scanning dependencies of target ncnnRay_lffd_face
[ 94%] Building CXX object CMakeFiles/ncnnRay_lffd_face.dir/src/ncnnRay_lffd_face.cpp.o
In file included from /src/ncnnRay/ncnnRay/src/ncnnRay_lffd_face.cpp:2:0:
/src/ncnnRay/ncnnRay/src/models/LFFD.h:6:1: warning: 'typedef' was ignored in this declaration
 typedef struct FaceInfo {

Compiling to WASM using Emscripten

  • Install emsdk
  • See web.bat (which is rather complicated to understand ...)
  • for Linux, you can use the provided docker file

Contributing

Feel free to report issues during build or execution. We also welcome suggestions to improve the performance of this application.

Citation

If you find the code or trained models useful, please consider citing:

@misc{ncnnRay++,
  author={Kashani, Shlomo},
  title={ncnnRay++2020},
  howpublished={\url{https://github.com/QuantScientist/ncnnRay/}},
  year={2020}
}

Disclaimers

  • No liability. Feel free to submit bugs or fixes.
  • No tech support: this is merely a spare-time fun project for me.
  • Contribution is more than welcomed though.
  • Tested only on Windows 10 with Visual Studio 2019. More OS and dev env support are welcomed.

Licensing

http://www.gnu.org/licenses/gpl-3.0.en.html

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of

MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

Third party licences:

  • raylib is licensed under an unmodified zlib/libpng license, which is an OSI-certified, BSD-like license that allows static linking with closed source software. Check LICENSE for further details.

  • NVIDIA CUDA license https://docs.nvidia.com/cuda/eula/index.html

References

About

ncnnRay++ is a CMake based integration of raylib and the very popular Tencent ncnn Deep Learning library. ncnn is written in C++ designed (but not only) for edge computing devices. The project depends on the Vulkan SDK (Vulakn is Khronos' API for Graphics and Compute on GPUs).

Resources

Readme

Code of conduct

Code of conduct

Security policy

Security policy
Activity

Stars

1 star

Watchers

2 watching

Forks

6 forks
Report repository

Releases

No releases published

Sponsor this project

Packages

No packages published

Languages