OpenCV + CUDA Docker Development Environment

A Docker-based C++ development environment with OpenCV (latest) and CUDA 13.0 support on Ubuntu 24.04. Includes GPU-accelerated computer vision, X11 display forwarding, and a simple build system via make.

Features

• Base image: nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04
• OpenCV: Built from source with full CUDA/cuDNN acceleration
• CUDA Compute: Configured for sm_89 (RTX 40-series); easy to adjust
• C++23 compilation via g++
• X11 forwarding for GUI windows (cv::imshow, etc.)
• Volume mount: host project directory mounted at /app inside the container
• FFmpeg / GStreamer / V4L2 codec support included

Requirements

Requirement	Notes
NVIDIA GPU	Compute capability ≥ 8.9 (or adjust CUDA_ARCH_BIN in Dockerfile)
NVIDIA driver	Compatible with CUDA 13.0
Docker Engine	20.10+
NVIDIA Container Toolkit	Enables --gpus all
16 GB free disk space	OpenCV source build is large
8 GB RAM	Recommended for parallel compilation

Project Structure

.
├── Dockerfile                  # Multi-layer image: CUDA base → deps → OpenCV build
├── docker-build.sh             # Build the Docker image
├── docker-run.sh               # Run container with GPU, X11, and volume mount
├── docker-clean.sh             # Remove the built Docker image
├── Makefile                    # Compile & run the test program inside the container
├── img/                        # Screenshots
│   └── sshot.jpg               # Screenshot of the running application
├── src/                        # Source code folder
│   └── test_opencv_cuda.cpp    # Sample program: draws text on an image, rotates using CUDA and shows on screen
└── build/                      # Compilation output (created by make)
    └── test_opencv_cuda        # Compiled binary

Screenshot

Quick Start

1. Build the Docker image

⚠️ This step compiles OpenCV from source and takes 20–60 minutes depending on your hardware.

./docker-build.sh

This builds and tags the Docker image as opencv-cuda-13.02-dev.

2. Start a development container shell

./docker-run.sh

This drops you into an interactive shell inside the container with:

• GPU access (--gpus all)
• Your project directory mounted at /app
• X11 display forwarding (for cv::imshow)

3. Build and run the sample program

Inside the container:

make        # compiles test_opencv_cuda.cpp → build/test_opencv_cuda
make run    # runs build/test_opencv_cuda (creates build/foo.jpg)
make clean  # removes build artifacts

The sample program:

1. Creates a 640×480 white image with a blue border and centered green "Hello World" text (CPU)
2. Uploads the image to the GPU (cv::cuda::GpuMat)
3. Rotates it 90° counter-clockwise using cv::cuda::rotate() — this confirms CUDA is working
4. Downloads the result back to CPU
5. Saves build/foo_rotated.jpg (rotated)
6. Displays finale image on host via X11

4. Clean up

To remove the Docker image when no longer needed:

./docker-clean.sh

Makefile Reference

Target	Description
make / make all	Compile test_opencv_cuda.cpp into build/test_opencv_cuda
make run	Run the compiled binary
make clean	Delete everything in build/

Compilation uses C++23 (-std=c++23 -O2) and links against OpenCV 4 (via pkg-config) and CUDA runtime.