OpenCV with CUDA in Docker

A Docker environment for building and running OpenCV 4.12.0 with CUDA/cuDNN GPU acceleration on Ubuntu 22.04.

---

What's Included

• OpenCV 4.12.0 built with opencv_contrib modules
• CUDA 12.2 + cuDNN 8 with DNN GPU backend enabled
• Hardware acceleration: OpenGL, GTK, GStreamer, FFmpeg, V4L2, dc1394, OpenEXR
• Dev tools: cmake, gcc/g++, clang, ninja, git, gdb, ccache, python3
• Non-root admin user with passwordless sudo

---

Requirements

• Host OS: Linux (tested on Ubuntu 22.04 / 24.04)
• NVIDIA GPU driver: version ≥ 535
• Docker Engine + NVIDIA Container Toolkit → Installation Guide

Check your driver version:

nvidia-smi

---

Getting Started

1. Build the Docker image

docker build -t opencv-gpu-build .

This will clone and compile OpenCV from source inside the image. It takes a while.

2. Run the container

Use the provided script and pass a local directory to mount as your workspace:

./run_dev.sh /path/to/your/project

This mounts your directory at /workspace inside the container and drops you into a bash shell.

Note: The script also forwards your display ($DISPLAY) and /dev/video0 for GUI windows and webcam access. Make sure X11 forwarding is allowed on the host if needed:

xhost +local:docker

3. Verify GPU access

Once inside the container, run:

python3 -c "import cv2; print('OpenCV:', cv2.__version__); print('CUDA devices:', cv2.cuda.getCudaEnabledDeviceCount())"

You should see your OpenCV version and a non-zero CUDA device count.

---

GPU Architecture

The Dockerfile defaults to CUDA architectures 86;89;90 (RTX 30/40 series). If your GPU is different, set the build arg:

docker build --build-arg CMAKE_CUDA_ARCHITECTURES=75 -t opencv-gpu-build .

Common values: 75 (RTX 20xx), 86 (RTX 30xx), 89 (RTX 40xx).

---

Building the Example Code

The opencv_example/ directory contains a YOLO11 object detection demo using OpenCV's DNN module.

Inside the container:

cd /workspace/opencv_example
mkdir build && cd build
cmake ..
make -j$(nproc)
./main

The example uses models/yolo11m.onnx (included) and the COCO class list from coco/coco.names.

---

Project Structure

.
├── Dockerfile                    # Builds the OpenCV + CUDA image
├── run_dev.sh                    # Helper script to run the container
├── opencv_example/               # YOLO11 object detection demo
│   ├── main.cpp
│   ├── CMakeLists.txt
│   ├── coco/                     # COCO class names and config
│   ├── models/                   # ONNX model files (yolo11m.onnx)
│   └── utils/                    # Helper scripts (pt_to_onnx, yaml_to_names)
└── opencv_lessons_code/          # Standalone lesson examples
    ├── lesson_1.cpp
    ├── lesson_2.cpp
    ├── histogram_calculation.cpp
    ├── shi-tomasi.cpp
    └── CMakeLists.txt

---

Lesson Code

The opencv_lessons_code/ directory contains standalone examples covering basic OpenCV GPU operations. Build them the same way as the main example:

cd /workspace/opencv_lessons_code
mkdir build && cd build
cmake ..
make -j$(nproc)