Skip to content

Latest commit

 

History

History
40 lines (26 loc) · 1.9 KB

README.md

File metadata and controls

40 lines (26 loc) · 1.9 KB
Raw

Docker Image ReadMe

The Dockerfile associated with this image can be used to build an image of the weld seam detection module that is waiting to receive a 3D mesh model of a welding scene from a server and a message to begin the the detection. When it is finished, it returns an array of welding paths.

Requirements

Usage

To use this image, a coordinator server to orchestrate the procedure is required. A server as such is provided in roboweldar.coordinator.

Deployment

To deploy the image, apart from Docker, the following are required:

  1. A computer that has an NVIDIA GPU with CUDA cores.

  2. CUDA Drivers. Tested with driver 440 - Cuda 10.2 and driver 450 - Cuda 11.

  3. NVIDIA docker. It can be downloaded by following the instructions from the official website or by running the quick installation script for Ubuntu.

How to build an image

The Dockerfile uses an NVIDIA Docker image as base, Alicevision's Merhroom runtime version 2019.2.0. While inside the root directory of the module, to build the image run

docker build -f docker/Dockerfile -t <component-name> .

How to run the image

Running the image requires to expose ports 3000 and 3001 to the network and the argument --host to show the IP of the server. The following are two sample cases:

  1. Running the image standalone:
docker run --runtime=nvidia -p 3000:3000 -p 3001:3001  -d <component-name> --host <server-ip>
  1. Running the image in the same host as the server and using the local network:
docker run --runtime=nvidia --network host -d <component-name> --host "localhost"