Using CUDA

Local installation

Prerequisites

To build/run locally you must install OpenCV with CUDA support.

For more information see:

https://github.com/hybridgroup/gocv/blob/release/cuda/README.md#installing-cuda

Building

Once you have installed the needed versions of CUDA and OpenCV, actually building the executable is much easier.

go install -tags cuda ./cmd/wasmvision

Running

wasmvision run -p mosaic --cuda-enable=true

Docker

You can also run a wasmVision Docker image with CUDA support.

Prerequisites

You must install the following software before running the wasmVision Docker image with CUDA support:

• CUDA version 12

• Nvidia Container Toolkit

Running

Pull the current development version:

docker pull ghcr.io/wasmvision/wasmvision-cuda-12:main

Verify it is installed like this:

docker run ghcr.io/wasmvision/wasmvision-cuda-12:main version

Now you can run a test to capture video using your webcam, blur it using a WebAssembly processor, and then stream the output to port 8080 on your local machine:

docker run --privileged --network=host --gpus all ghcr.io/wasmvision/wasmvision-cuda-12:main run -p /processors/mosaic.wasm --cuda-enable=true

Point your browser to http://localhost:8080 and you can see the output.