docker-node-opencv-cuda

Features

Working OpenCV+CUDA Docker image for your Node.js applications.

• powerful with GPU support: Built on the latest official CUDA images to enable CUDA-powered GPU acceleration.

• streamlined: Node is installed exactly like it is in the official Node.js Dockerfile to enable out-of-the box support for application targeting existing node images.

• lightweight: We are using Multi-stage builds to efficiently switch to a devel image used to build opencv to a lighter runtime image for the final release.

• modern: We are targeting the latest OpenCV releases and plan to keep it up-to-date as much as possible.

Getting started

Check the available tags on dockerhub.

• Using the command line interface:

docker pull mgcrea/node-opencv-cuda:18-opencv-4-cuda-12-cudnn8-runtime-ubuntu22.04
docker run -it --rm --runtime=nvidia --gpus all mgcrea/node-opencv-cuda:18-opencv-4-cuda-12-cudnn8-runtime-ubuntu22.04 nvidia-smi

• Using your own Dockerfile (basic example):

FROM mgcrea/node-opencv-cuda:18-opencv-4-cuda-12-cudnn8-runtime-ubuntu22.04

# Install app
WORKDIR /app
COPY package.json .
COPY build ./build/
COPY node_modules ./node_modules/

EXPOSE 3000

CMD node .

Notes

To fully leverage your GPU capabilities, you might need to tweak the CUDA_ARCH_BIN according to your GPU, you should check your GPU Compute Capability and build this image accordingly:

docker build --build-arg CUDA_ARCH_BIN=8.6 .

Multiple architectures can be supported with comma-separated values (eg. CUDA_ARCH_BIN=8.0,8.6).