[Deep Learning] User Guide for Common Dataset Labeling Software

Article directory

foreword

When we are doing instance segmentation or target detection related projects, we usually label our data sets according to the algorithm requirements. The labeling formats are usually json, xml, txt and other formats. Comrades who are not familiar with these three common labeling formats You can move to: Convert the VOC (xml) annotation format to YOLOv5 (txt) and COCO2017 (json) formats .

This paper mainly introduces three kinds of data annotation software commonly used in the field of instance segmentation and object detection.

Instance segmentation: Labelme

Anaconda Prompt Installation

conda create -n labelme python=3.8
conda activate labelme
pip install pyqt5 # pyqt5 can be installed via pip on python3
pip install labelme==3.16.2

Annotation step

Open the software

Enter the labelme environment in Anaconda Prompt

activate labelme

cd to the specified folder
then run the labelme command
- --labels: pass in all the defined class names
- –output: Specify the save address of the annotation file

(labelme) D:\dataset>labelme --labels label.txt --output output

Bulk annotation

Click Open Dirto open the folder where the dataset is located
Click Create Polygonsto create a polygon labeling form and start labeling
Choose a label category
Click Saveto save the annotation file of the image (json format)
Click Next Imageto mark the next picture

insert image description here

Object detection: LabelImg

Anaconda Prompt Installation

conda create -n labelimg python=3.8
conda activate labelimg
pip install labelimg

Annotation step

Open the software

Enter the labelimg environment in Anaconda Prompt

activate labelimg

cd to the specified folder
Then run the labelimg command
- images: folder where images are stored
- labels.txt: specifies all the categories to be labeled

(labelimg) D:\dataset>labelimg images labels.txt

Settings before labeling
Click Viewto display the image below, and then tick the following options:

Auto Save mode : When switching to the next picture, the label of the previous marked picture will be automatically saved, so that you don't need to press Ctrl+S to save each marked picture.
Display Labels : After marking the picture, the box and label will be displayed
Advanced Mode : In this way, the marked cross will always be suspended in the window, without having to mark a target each time, press the W shortcut key again to call up the marked cross

Common shortcut keys

W: Call out the marked cross and start marking
A: Switch to the previous picture
D: switch to the next picture
Ctrl+S: Save the marked label
del: delete the marked rectangle
Ctrl+Mouse Wheel: Press and hold Ctrl, then scroll the mouse wheel to adjust the display size of the marked image
Ctrl+u: Select the folder where you want to annotate pictures
Ctrl+r: Select the folder where the labeled label is stored
↑→↓←: Move the position of the marked rectangle

Bulk annotation

Click Open Dirto open the folder where the dataset is located
Click Create RectBoxto create a rectangular callout box and start marking
Choose a label category
Select the label format, there are three label formats in labelimg, you can choose one:
- Pascal VOC(xml)
- YOLO(txt)
- CreateML(json)
Click Saveto save the annotation file of the image
Click Next Imageto mark the next picture

Instance Segmentation & Object Detection: Colabeler

Compared with Labelme, LabelImg these annotation tools, the powerful part of the wizard annotation assistant is that it supports instance segmentation, target detection, text annotation, audio annotation and video annotation, etc. , and it is completely free, which can be called the conscience of the industry!

Elf Marking Assistant currently supports Windows/Mac/Linux platforms, and you can download the corresponding version according to your own system.

Download directly from the official website

Colabeler

Labeling step (take target detection as an example)

New Project

Open the software, complete the registration and login, click New in the upper left corner, you can see that there are many types of projects supported, here select the first position to mark
Then fill in the relevant information on the right and click Create

insert image description here

Bulk annotation

There are three types of callout boxes that can be selected. Here, select the rectangle box to start marking.
Then select the label information in the upper right corner
Click the checkmark √ below, orCtrl+s
Then you can click the previous one on the left, or directly use the left and right buttons of the keyboard to switch pictures

insert image description here