labelme: Image Annotation Tool with Python

Labelme is a graphical image annotation tool inspired by http://labelme.csail.mit.edu.
It is written in Python and uses Qt for its graphical interface.

Requirements

Ubuntu / macOS / Windows
Python2 / Python3
PyQt4 / PyQt5

Installation

There are options:

Platform agonistic installation: Anaconda, Docker
Platform specific installation: Ubuntu, macOS

Anaconda

You need install Anaconda, then run below:

conda create --name=labelme python=2.7
source activate labelme
conda install pyqt
pip install labelme

Docker

You need install docker, then run below:

wget https://raw.githubusercontent.com/wkentaro/labelme/master/scripts/labelme_on_docker
chmod u+x labelme_on_docker

# Maybe you need http://sourabhbajaj.com/blog/2017/02/07/gui-applications-docker-mac/ on macOS
./labelme_on_docker static/apc2016_obj3.jpg -O static/apc2016_obj3.json

Ubuntu

sudo apt-get install python-qt4 pyqt4-dev-tools
sudo pip install labelme

macOS

brew install qt qt4 || brew install pyqt  # qt4 is deprecated
pip install labelme

Usage

Annotation

Run labelme --help for detail.

labelme  # Open GUI
labelme tutorial/apc2016_obj3.jpg  # Specify file
labelme tutorial/apc2016_obj3.jpg -O tutorial/apc2016_obj3.json  # Close window after the save

The annotations are saved as a JSON file. The file includes the image itself.

Visualization

To view the json file quickly, you can use utility script:

labelme_draw_json tutorial/apc2016_obj3.json

Convert to Dataset

To convert the json to set of image and label, you can run following:

labelme_json_to_dataset tutorial/apc2016_obj3.json -o tutorial/apc2016_obj3_json

It generates standard files from the JSON file.

img.png: Image file.
label.png: Int32 label file.
label_viz.png: Visualization of label.png.
label_names.txt: Label names for values in label.png.

Note that loading label.png is a bit difficult (scipy.misc.imread, skimage.io.imread may not work correctly), and please use PIL.Image.open to avoid unexpected behavior:

# see tutorial/load_label_png.py also.
>>> import numpy as np
>>> import PIL.Image

>>> label_png = 'tutorial/apc2016_obj3_json/label.png'
>>> lbl = np.asarray(PIL.Image.open(label_png))
>>> print(lbl.dtype)
dtype('int32')
>>> np.unique(lbl)
array([0, 1, 2, 3], dtype=int32)
>>> lbl.shape
(907, 1210)

Screencast

Acknowledgement

This repo is the fork of mpitid/pylabelme, whose development has already stopped.