Adapted LabelImg for Enhanced User Experience

This repository is a copy from HumanSignal/labelImg. The original repository is archived and no longer being maintained. So I make a copy from the latest version (1.8.6) to modify some function and fix some error for personal use.

Original README

Installation

Install with pip

pip install -U mlabelImg

Install from source

git clone https://github.com/PD-Mera/mlabelImg
pip install pyqt5 lxml
pyrcc5 -o mlabelImg/libs/resources.py mlabelImg/resources.qrc
pip install -e mlabelImg

Usage

Setup directory

Create a folder structure same as below

├── data
    ├── images
    └── labels

Put all of your image in images directory. And create a classes.txt contain all class you want to label. Example of classes.txt as below

dog
cat
pig

Put classes.txt in 2 place, in labels directory and same level as labels directory

Full structure of workspace as below

├── data
    ├── images
    │   ├── img1.jpg
    │   ├── img2.jpg
    │   └── ...
    ├── labels
    │   └── classes.txt
    └── classes.txt

Run mlabelImg

Run mlabelImg with

# mlabelImg [IMAGE_PATH] [PRE-DEFINED CLASS FILE]
mlabelImg .\data\images\ .\data\classes.txt

On GUI of labelImg:

• File -> Change Save Dir -> (save label directory)
• Choose YOLO format on the left tray

Next and previous image with D -> A

Label with W

Delete .\data\classes.txt after labeling

Label format

With YOLO format, label will be saved with format label_index x_center y_center w h and normalize to scale [0, 1]

1 0.415842 0.863095 0.102970 0.101190
1 0.228713 0.315476 0.077228 0.053571
1 0.756436 0.328869 0.114851 0.050595

Reference

• Author: TzuTa Lin
• Author Email: tzu.ta.lin@gmail.com
• tzutalin/labelImg