mrjjq/gpupixel

Real-Time video and image AI beauty filter library that achieves commercial-grade beauty effects. It is written in C++11 and based on OpenGL/ES.

English | 简体中文

---

Welcome to join us to make GPUPixel better by participating discussions, opening issues, submitting PRs.

📢 The face landmark detection has replaced Face++ with VNN starting from v1.2.0; no more Face++ network verification requirements are needed 👏

Introduction

⛰️ GPUPixel is a real-time, high-performance image and video filter library, extremely easy to compile and integrate with small library size.

🔑 GPUPixel is written in C++11 and is based on OpenGL/ES, incorporating a built-in beauty face filter that achieves commercial-grade beauty effects.

🔌 GPUPixel supports platforms including iOS, Android, Mac, Win and Linux. It can theoretically be ported to any platform that supports OpenGL/ES.

Effects Preview

👉 Video: YouTube | BiliBili

Origin	Smooth	White	ThinFace
BigEye	Lipstick	Blusher	ON-OFF

Architecture

Features

This table compares the features supported by GPUPixel and GPUImage and Android-GPUImage:

✅: Supported | ❌: Not supported | ✏️: Planning

	GPUPixel	GPUImage	Android-GPUImage
📷 Filters:	✅	❌	❌
Skin Smoothing Filter	✅	❌	❌
Skin Whitening Filter	✅	❌	❌
Face Slimming Filter	✅	❌	❌
Big Eyes Filter	✅	❌	❌
Lipstick Filter	✅	❌	❌
Blush Filter	✅	❌	❌
More Build in Filter	✅	✅	✅
🎬 Input Formats:
YUV420P(I420)	✅	❌	❌
RGBA	✅	✅	✅
JPEG	✅	✅	✅
PNG	✅	✅	✅
NV21(for Android)	✏️	❌	❌
🎥 Output Formats:
RGBA	✅	✅	✅
YUV420P(I420)	✏️	❌	❌
💻 Platform:
iOS	✅	✅	❌
Mac	✅	✅	❌
Android	✅	❌	✅
Win	✅	❌	❌
Linux	✅	❌	❌

Performance

iPhone

-	iPhone 6P	iPhone 8	iPhone X	iPhone 11	iPhone 14 pro
CPU	5%	5%	3%	3%	3%
Time Taken	10ms	4ms	3ms	3ms	3ms

Android

-	Xiaomi 10	Huawei Mate30	Vivo	SAMSUNG	Google Pixel
CPU	3%	5%	-	-	-
Time Taken	6ms	5ms	-	-	-

Lib Size

	iOS(.framework)	MacOS(.framework)	Android(.aar)
Size	2.4 MB	2.6 MB	2.1 MB

Before You Start

⭐️ Star us on GitHub, and be instantly notified for new releases!

Getting Started

How to Build

Compile using CMake frome v1.1.0. Please search for instructions on how to install and configure CMake.

The generated libraries and app will be located in the "output" directory of the project's root folder.

iOS

cd src
mkdir build
cd build

# Generate project
## for iOS arm64
cmake -G Xcode -DCMAKE_TOOLCHAIN_FILE=../../toolchain/ios.toolchain.cmake -DPLATFORM=OS64 ..
 
# Build
cmake --build . --config Debug

Mac

cd src
mkdir build
cd build

# Generate project
## for Mac Apple Silicon
cmake -G Xcode -DCMAKE_TOOLCHAIN_FILE=../../toolchain/ios.toolchain.cmake -DPLATFORM=MAC_ARM64 ..
## for Mac Intel
cmake -G Xcode -DCMAKE_TOOLCHAIN_FILE=../../toolchain/ios.toolchain.cmake -DPLATFORM=MAC ..

# Build
cmake --build . --config Debug

Android

Open the directory src/android/java in Android Studio.

Windows

You need install and config Cmake and MinGW64 by yourself.

cd src
mkdir build
cd build

# Generate project
cmake -G "MinGW Makefiles" ..

# Build
mingw32-make

Linux (Test on ubuntu)

# install cmake 
sudo apt-get install cmake pkg-config
# install dependent lib
sudo apt-get install mesa-utils libglu1-mesa-dev freeglut3-dev mesa-common-dev libglfw3-dev

# start build
cd src
mkdir build
cd build

# Generate project
cmake ..

# Build
make

App demo

iOS and Mac

Refer to examples/ios and examples/mac

Android

Refer to examples/android or src/android/java

Window and Linux

Refer to examples/desktop . The compilation method is the same as compiling the library.

cd examples
mkdir build
cd build

# Generate project
cmake -G "MinGW Makefiles" ..

# Build
mingw32-make

App Usage

A S D F G H - Increase smooth, white, thin face, big eye, lipstick, blusher level

Z X C V B N - Decrease smooth, white, thin face, big eye, lipstick, blusher level

How to Use

Declear filters

// video data input
std::shared_ptr<SourceRawDataInput> source_raw_input_;
// beauty filter
std::shared_ptr<BeautyFaceFilter> beauty_face_filter_;
// video data output 
std::shared_ptr<TargetRawDataOutput> target_raw_output_;

Create and link filters

 gpupixel::GPUPixelContext::getInstance()->runSync([&] {
    // Create filter
    source_raw_input_ = SourceRawDataInput::create();
    target_raw_output_ = TargetRawDataOutput::create();
    // Face Beauty Filter
    beauty_face_filter_ = BeautyFaceFilter::create();
    
    // Add filter
    source_raw_input_->addTarget(beauty_face_filter_)
                     ->addTarget(target_raw_output_);
 }

Input YUV420P or RGBA

// ...
// YUVI420
 source_raw_input_->uploadBytes(width,
                                height, 
                                bufferY,
                                strideY, 
                                bufferU, 
                                strideU,
                                bufferV, 
                                strideV);
// ...
// bytes: RGBA data
 source_raw_input_->uploadBytes(bytes,
                                width, 
                                height, 
                                stride);

Output Data Callback

// I420 callback
target_raw_output_->setI420Callbck([=](const uint8_t *data, 
                                        int width, 
                                        int height, 
                                        int64_t ts) {
    size_t y_size = width * height;
    const uint8_t *uData = data + y_size;
    const uint8_t *vData = data + y_size + y_size / 4;
    // Do something you want
});

// RGBA callback->
target_raw_output_->setPixelsCallbck([=](const uint8_t *data, 
                                        int width, 
                                        int height, 
                                        int64_t ts) {
    size_t rgba_size = width * height*4;
    // Do something you want
});

// Output data callbck

Star History

Contributing

Welcome to contribute code 👏🏻.

At the same time, please consider supporting GPUPixel by sharing it on social media and at events and conferences.

Acknowledgement

Reference Project

1. GPUImage
2. CainCamera
3. AwemeLike
4. VNN

License

This repository is available under the MIT License.