GCV

OpenGL/OpenCL/VAAPI interop demos (aka. run it on the GPU!) using my 4.x fork of OpenCV (https://github.com/kallaballa/opencv/tree/GCV)

The goal of the demos is to show how to use OpenCL interop in conjunction with OpenCV on Linux to create programs that run mostly (the part the matters) on the GPU. Until the necessary changes are pulled into the official repository you need to build my fork of OpenCV 4.x.

• The author of the example video (which is also used for two of the demo videos in this README) is (c) copyright Blender Foundation | www.bigbuckbunny.org.
• The author of the video used for pedestrian detection is GNI Dance Company (Original video)
• The right holders of the video used for the optical flow visualization are https://www.bbtv.com. I tried to contact them several times to get an opinion on my fair-use for educational purpose. (Original video)

Requirements

• Support for OpenCL 1.2
• Support for cl_khr_gl_sharing and cl_intel_va_api_media_sharing OpenCL extensions.
• If you are on a recent Intel Platform (Gen8 - Gen12) you need to install an alternative compute-runtime

Dependencies

• EGL
• GLEW
• GLFW3
• nanovg
• nanogui

There are currently seven demos but only two are maintained till I am done with the API redesign (the preview videos are scaled down and compressed):

video-demo

Renders a rainbow tetrahedron on top of a input-video using OpenGL, applies a glow effect using OpenCV (OpenCL) and decodes/encodes on the GPU (VAAPI).

https://user-images.githubusercontent.com/287266/205186059-c96e2728-62e4-41c5-b6a2-c7e393efeda2.mp4

optflow-demo

My take on a optical flow visualization on top of a video. Uses background subtraction (OpenCV/OpenCL) to isolate areas with motion, detects features to track (OpenCV/OpenCL), calculates the optical flow (OpenCV/OpenCL), uses nanovg for rendering (OpenGL) and post-processes the video (OpenCL). Decodes/encodes on the GPU (VAAPI).

https://user-images.githubusercontent.com/287266/202174513-331e6f08-8397-4521-969b-24cbc43d27fc.mp4

Instructions

You need to build my 4.x branch of OpenCV, nanovg and nanogui.

Build nanovg

git clone https://github.com/inniyah/nanovg.git
mkdir nanovg/build
cd nanovg/build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j8
sudo make install

Build nanogui

git clone https://github.com/mitsuba-renderer/nanogui.git
mkdir nanogui/build
cd nanogui/build
cmake -DCMAKE_BUILD_TYPE=Release -DNANOGUI_BACKEND=OpenGL -DNANOGUI_BUILD_EXAMPLES=OFF -DNANOGUI_BUILD_GLFW=OFF -DNANOGUI_BUILD_PYTHON=OFF ..
make -j8
sudo make install

Build OpenCV

git clone --branch GCV https://github.com/kallaballa/opencv.git
cd opencv
mkdir build
cd build
ccmake -DCMAKE_BUILD_TYPE=Release -DOPENCV_ENABLE_GLX=ON -DOPENCV_ENABLE_EGL=ON -DOPENCV_FFMPEG_ENABLE_LIBAVDEVICE=ON -DWITH_OPENGL=ON -DWITH_QT=ON -DWITH_FFMPEG=ON -DOPENCV_FFMPEG_SKIP_BUILD_CHECK=ON -DWITH_VA=ON -DWITH_VA_INTEL=ON -DBUILD_PERF_TESTS=OFF -DBUILD_TESTS=OFF -DBUILD_EXAMPLES=OFF ..
make -j8
sudo make install

Build demo code

git clone https://github.com/kallaballa/GCV.git
cd GCV
make

Download the example file

wget -O bunny.webm https://upload.wikimedia.org/wikipedia/commons/transcoded/f/f3/Big_Buck_Bunny_first_23_seconds_1080p.ogv/Big_Buck_Bunny_first_23_seconds_1080p.ogv.1080p.vp9.webm

Run the video-demo:

src/video/video-demo bunny.webm

Run the optflow-demo:

src/optflow/optflow-demo bunny.webm