1. What is labelme

2. Detailed installation tutorial

1. What is labelme

labelme is a graphic image annotation tool, it is written in Python, in fact, it is a labeling software,

--- Points, rectangles, polygons, etc. can be used for target detection and segmentation tasks, etc.

---Annotation in the form of flag, which can be used for image classification and cleaning tasks

---Video annotation, generate VOC data set

---Generate a dataset in COCO format

2. Detailed installation tutorial①

Reference link https://github.com/wkentaro/labelme

First, ancoda must be installed, and the specific installation process will not be repeated here.

Click Anaconda Prompt to install. All commands are performed here.

① Check the python version code

python --version

②Create a conda environment for labelme, named labelme

conda create --name=labelme python=3.7

Because I have already created it, it is in this state. If you are connected to the external network, be careful to disconnect the external network.

③ Activate the environment

conda activate labelme

The front end will change. operate in this environment.

④Install labelme

pip install labelme

Appearance of successful installation

 

For the next use, activate labelme and enter labelme directly.

Interface description

 

⑤ You can create labels with commands, or you can create labels yourself. Here is the way to create your own labels.

Create a text document txt file in the same directory as the image folder to be marked,Note that txt must be created in the same directory

 

 

The first two are a must. __ignore__ and _background_ must have, below is the label. The underline in front of ignore is to enter two underscores consecutively.

 

⑥ Take segmentation as an example

Open this file path in the terminal with .

cd command

cd .. #return to the previous directory

cd \ # Return to the root directory under the current directory

cd file or folder path #Access the path of the c drive

E:+Enter+cd command #To access the path of the E drive, first enter the drive letter, press Enter, and then use the cd command

Type in the terminal

labelme data_annotated --labels labels.txt --nodata --validatelabel exact --config "{shift_auto_shape_color: -2} "

 data_annotated is the name of the image folder to be labeled, and labels.txt is the txt file created above.Note that there is a single quote at the end on github, but it needs to be changed to double quotes in windows.

 

 

⑦File format conversion

Since the marked file is in json format, we will convert it into image format or other formats. Use the code labelme2voc.py given by the author. The main conversion formats are: class name file, jpeg format of the original image, npy format of the segmented image, png format, and jpg format of overlaying the segmented image on the original image

Author's order:

./labelme2voc.py data_annotated data_dataset_voc --labels labels.txt

But it will report an error when running, modify it to the following code

python labelme2voc.py data_annotated data_dataset_voc --labels labels.txt

 data_annotated is the name of the annotated image, and data_dataset_voc is the name of the newly created VOC format file.

existThe results of label segmentation can be viewed in data_dataset_voc.