Skip to content

Latest commit

 

History

History
275 lines (204 loc) · 9.06 KB

cvat-cnv.md

File metadata and controls

275 lines (204 loc) · 9.06 KB
Raw

MANUela Logo

Deploy and demo CVAT on OpenShift Virtualization

This document describes how to deploy and demonstrate CVAT on OpenShift Virtualization

Prerequisites

  • OpenShift Virtualization is deployed and configured on you target cluster
  • virtctl on your laptop (install demo only)
  • Project manuela-visual-inspection has been created.

Install CVAT in a OpenShift Virtualization virtual machine

Install CentOS Steams VM

In the OpenShift Console go into the manuela-visual-inspection project and navigate to Virtualization -> VirtualMachines.

  • Select Create -> From catalog -> CentOS Stream 8 VM -> Customize VirtualMachine
  • Enter General data:
    • Name: cvat
    • Optional parameter CLOUD_USER_PASSWORD: redhat
  • Review and Create Virtual Machine
  • Start Virtual Machine
  • Check the Console and login

Install CVAT

The steps below are based on CVAT Quick installation guide

Login into the CentOS Streams VM, install and start tmux

# oc get vmi
NAME   AGE   PHASE     IP             NODENAME
cvat   13d   Running   10.131.0.109   storm5-10g.ocp5.stormshift.coe.muc.redhat.com

# virtctl console cvat
Successfully connected to cvat console. The escape sequence is ^]

[centos@cvat ~]$ sudo dnf install tmux -y
[centos@cvat ~]$ tmux

Install Docker and Compose

# sudo systemctl disable firewalld
# sudo dnf install iptables -y

# sudo dnf config-manager --add-repo=https://download.docker.com/linux/centos/docker-ce.repo
# sudo  dnf install docker-ce --nobest -y
# sudo systemctl enable --now docker

# systemctl is-active docker
active
# systemctl is-enabled docker
enabled

# sudo docker --version
Docker version 20.10.3, build 48d30b5

Enable centos user

# sudo usermod -aG docker centos

Log out and log back in (or reboot) so that your group membership is re-evaluated.

Test Docker

# docker run hello-world

Install Compose

# curl -L https://github.com/docker/compose/releases/download/1.25.4/docker-compose-`uname -s`-`uname -m` -o docker-compose
# sudo mv docker-compose /usr/local/bin && sudo chmod +x /usr/local/bin/docker-compose

Install CVAT via compose

# sudo dnf install git -y
# git clone https://github.com/opencv/cvat
# cd cvat
# git checkout tags/v1.4.0
# vim docker-compose.yml

Add the proper image tags to the CVAT containers:

...
cvat:
  container_name: cvat
  image: openvino/cvat_server:v1.4.0
...
cvat_ui:
  container_name: cvat_ui
  image: openvino/cvat_ui:v1.4.0
...

Then spin up the containers:

# docker-compose up -d

Create superuser

I.e. django / password

# docker exec -it cvat bash -ic 'python3 ~/manage.py createsuperuser'
Username (leave blank to use 'django'): 
Email address: 
Password: password
Password (again): password
This password is too common.
Bypass password validation and create user anyway? [y/N]: y
Superuser created successfully.

Check container

# docker ps
CONTAINER ID   IMAGE                  COMMAND                  CREATED         STATUS         PORTS                  NAMES
8301ab32e473   nginx:stable-alpine    "/docker-entrypoint.…"   7 minutes ago   Up 7 minutes   0.0.0.0:8080->80/tcp   cvat_proxy
4ba51027a506   openvino/cvat_ui       "/docker-entrypoint.…"   7 minutes ago   Up 7 minutes   80/tcp                 cvat_ui
accf8d79a21c   openvino/cvat_server   "/usr/bin/supervisord"   7 minutes ago   Up 7 minutes   8080/tcp               cvat
6d255df66037   redis:4.0-alpine       "docker-entrypoint.s…"   7 minutes ago   Up 7 minutes   6379/tcp               cvat_redis
20fc714a68e6   postgres:10-alpine     "docker-entrypoint.s…"   7 minutes ago   Up 7 minutes   5432/tcp               cvat_db

Check if the CVAT web site is responding

# curl localhost:8080
...
        <meta
            name="description"
            content="Computer Vision Annotation Tool (CVAT) is a free, open source, web-based image and video annotation tool which is used for labeling data for computer vision algorithms. CVAT supports the primary tasks of supervised machine learning: object detection, image classification, and image segmentation. CVAT allows users to annotate data for each of these cases"
        />
        <meta name="”robots”" content="index, follow" />
        <title>Computer Vision Annotation Tool</title>
...

Expose VCAT externally


Replace `.apps.ocp5.stormshift.coe.muc.redhat.com` with your `CVAT_HOST`
# docker-compose down

# cat <<EOF > docker-compose.override.yml 
version: '3.3'

services:
  cvat_proxy:
    environment:
      CVAT_HOST: .apps.ocp5.stormshift.coe.muc.redhat.com
EOF

# docker-compose up -d

Exit VM shell

exit

Expose port as service and create route

# virtctl expose virtualmachineinstance cvat --port=8080 --name=cvat
# oc create route edge --service=cvat

# oc get route cvat
NAME   HOST/PORT                                                                PATH   SERVICES   PORT    TERMINATION   WILDCARD
cvat   cvat-manuela-visual-inspection.apps.ocp5.stormshift.coe.muc.redhat.com          cvat       <all>   edge          None

Add User in CVAT

Open your CVAT Django Admin page and add a user. E.g.,

https://cvat-manuela-visual-inspection.apps.ocp5.stormshift.coe.muc.redhat.com/admin/

Django Admin

Add User. E.g. manuela / cvatdemo Add User

Add user to annotator, observer and user group Add Group

Save and logout.

For further details see the CVAT Users Guide

Image labeling with the Computer Vision Annotation Tool

The demo has multiple parts and you can pick and choose depending on you audience.

  1. Show Docker application running in OpenShift Virtualization VM
  2. Demo the annotation workflow for the visual inspecting showcase

Show Docker application running in OpenShift Virtualization VM

For a long demo, walk through the installation above. Alternatively, you can just open the console of the cvat virtual machine, login and show the running container with docker ps: CVAT VM

Additional, navigate to the routes and open the cvat route.

Demo the annotation workflow for the visual inspecting showcase

Show unlabeled images on file systems

With your favorite file explorer, navigate to manuela-visual-inspection/cam/data/metal_nut/scratch/ and filter for png files and explain the data: List of scratch bent and good images Unlabled Images

Upload and show images in CAVT

  • Open CVAT in E.g. https://cvat-manuela-visual-inspection.apps.ocp5.stormshift.coe.muc.redhat.com
  • Login as manuela user.
  • Create a new project demo and add the labels scratch and bent to the project: Create Project
  • Open the demo project and create a new task: Create Task
  • Name the task metal-nut and upload data.zip images from manuela-visual-inspection/ml/pytorch/ using Click or drag files to this area: Upload Data
  • Submit task. The data are being uploaded to the server..
  • Navigate to Tasks: Tasks

Annotate Showcase manual annotation:

  • Open Job #1: Open Job
  • Show how to annotate manually: annotate

Upload pre-labeled annotations:

  • Navigate to the Task list and select Actions -> Upload Annotations -> YOLO 1.1
  • Upload annotation file metal-nut-annotations.zip from manuela-visual-inspection/ml/pytorch/.
  • In case the warning Current annotation will be lost appears, press Update.

Show pre-labeled data:

  • Open the task and job
  • Show annotations annotations
  • The dataset is ready to training!

Demo Clean-up

Clean-up annotation workflow

  • In CVAT, navigate to projects and delete demo project.

Delete cvat VM

  • Delete VM, DV, Service and Route
oc project manuela-visual-inspection
oc delete vm cvat
oc delete dv centos8-stream
oc delete route cvat
oc delete service cvat