gimp-image-annotator
gimp-image-annotator or GIÀ, a lightweight GIMP plug-in to alllow for computer vision-assisted image annotation using the powerful GIMP selection toolbox.
Installation
Follow the guide here: https://en.wikibooks.org/wiki/GIMP/Installing_Plugins to find how to install GIMP plug-ins on your system, save the file image-annotator.py
in GIMP's plug-in folder.
In GIMP v2.x, the plug-in system relies on deprecated python2. On Windows, a version of python2 is included in the installation of GIMP, so you only need to follow the plug-in installation. On Linux, we recommend using the Flatpak version of GIMP, as it comes with the correct python2 binaries inlcluded. On Linux, the plug-in may need to be made executable with the command chmod a+x /path/to/image-annotator.py
in order to be seen by GIMP.
Using the software Once installed, navigate to Toolbox then Image Annotator, add the labels you want, select one, use GIMP's selection tools (e.g. The Fuzzy Select tool - a guide can be found here: https://docs.gimp.org/en/gimp-tools-selection.html) to select an area (use Quick Mask or Shift+Q to quickly see the mask you have created). Make sure antialisaing and feathering is off, you cannot turn it off for rectangle select however it isn't used. Once you have your desired selected area, press Save selected mask. Repeat until all objects are annotated.
How do I use the data?
gimp-image-annotator saves a binary mask of each annotation, with class of mask stored in the _annotations.json
file. The _annotations.json
file is structured as followed
[{ "label": "label", "id": "0", "filename": "image.png" }]
The masks can be inputted using most image processing software. For example in opencv-python
it would be:
import cv2; mask = cv2.imread(PATH, cv2.IMREAD_GRAYSCALE)
Version History
main @ 0b0f345 (earliest) Created 13th Dec 2024 at 11:15 by Kieran Atkins
Update README.md
Frozen
main
0b0f345
Creator
Submitter
Views: 204 Downloads: 28
Created: 13th Dec 2024 at 11:15
Last updated: 13th Dec 2024 at 11:44
This item has not yet been tagged.
None