COCO-Annotator installation and use [very detailed]

COCO-Annotator installation and use

1. Required environment

Docker needs to be installed before installing COCO-Annotator. If necessary, you can refer to this article: Install docker under Windows

2. Install

  1. Enter cmd in the directory to be installed
  2. input the command
git clone https://github.com/jsbroks/coco-annotator
  1. cd command to enter the folder
cd coco-annotator
  1. open docker
  2. Enter the command and wait
docker-compose up

The download is relatively slow, and you can change to a domestic image (although I said so, but it is not much faster when I change to a domestic image), because I downloaded the Windows desktop version of docker, so click Settings in the interface ->Docker Engine, change the code on the right to the following code:

{
  "builder": {
    "gc": {
      "defaultKeepStorage": "20GB",
      "enabled": true
    }
  },
  "experimental": false,
  "features": {
    "buildkit": true
  },
  "registry-mirrors": [
    "https://registry.docker-cn.com",
    "http://hub-mirror.c.163.com",
    "https://docker.mirrors.ustc.edu.cn"
  ]
}

3. Use

After downloading, the service will be started automatically

Enter http://localhost:5000 in the browser

3.1 Create user

The first time you log in, you need to register an account

For the account registered for the first time, the user authority is automatically set to Admin

User rights:

  • Admin: the user with the highest authority
  • Owner: the creator of a particular dataset
  • User: The user who shared the dataset with Admin or Owner

Different user permissions are as follows:

Permission User Owner Admin
Browse pictures
Create/Edit/Delete Dimensions
Import Dimensions
Share/download/delete dataset
Add/download/delete pictures
Add picture
Edit other users’ datasets

Admin->Create User Create a new user

Here I created three users

3.2 Create category

After the registration is complete, click Categories->Create

Here I have created a category of 17 human body key points in MS COCO format. It is recommended to write all the Label first and then fill in Connects to

3.3 Create dataset

createCreate a dataset

Fill in the name of the data set and the classification in the data set, and create a new data set test_1

Click on the newly created dataset test_1, you can see that there is nothing in it

Enter the directory where you installed coco-annotator. The datasets folder in the directory is the dataset you created. Here is D:\software\coco-annotator\datasets\test_1, after moving the picture to this directory, it is still blank. At this point, click the Scan button on the left column and refresh.

3.4 Annotating images

  1. First click + on the right
  2. Select BBox Tool

  1. Use BBox to select people first
  2. Click ④ in the right column
  3. click nose

  1. Start marking from nose, just click on the picture

The labeling results are as follows

There is still a small problem here. From the picture, you can see that the face and right ear of the person on the right are not exposed. They are invisible to those who look at the picture. Here we take the nose as an example, click the small gear button

Change the visibility of the nose to LABELED_NOT_VISIBLE, then click close to close

If the corresponding point does not appear in the figure, there is no need to label

3.5 Export annotation information

3.5.1 Export all image annotations

Click Export COCO on the left navigation bar, the following page will pop up, select the category to be exported, and click Export to export. Then, click Exports in the upper navigation bar, and the marked download information will appear, click Download to download.

3.5.2 Export single image annotation

To export a single image annotation, just click Download Image & amp; COCO

3.6 Remote Access

3.6.1 Enable remote access

Open the Control Panel, enter the following path, and click Advanced Settings

Click Inbound Rules->New Rule->Port

Then go to the next step directly to the end, confirm the name, and click Finish

3.6.2 Remote access

Use the ipconfig command to check your own IP address, and change the localhost in the original access address http://localhost:5000/ to your own IP address. If there is a public network IP, it can be accessed from the external network. If it is a local area network IP, it can only be accessed within the local area network. If you want to access from the external network, you need to perform IP address mapping.

ps. Supplement the user section

Log in to the current admin account to see

Log in to user1 and there is nothing

In the admin account, click the data set to be collaborated and select Share

Add user1 and save

At this time, go back to user1, refresh and you can see the data set shared by the admin account, and you can mark it together

References

  1. Installation and use of coco-annotator
  2. COCO dataset Keypoint annotation format combing and using COCO-Annotator to annotate on its own dataset
  3. Usage·jsbroks/coco-annotator-Wiki