🐦 multi_label_pigeon

This repo is a simple multiclass image extenstion of the pigeon library by @agermanidis located here. Credit goes to them for building the annotate function for images, text, and regression for multi-class type problems. All I am doing is extending it to multi-label and multi-task type problems.

Installation

Can install with standard pip.

pip install multi-label-pigeon

Examples

from multi_label_pigeon import multi_label_annotate
from IPython.display import display, Image

annotations = multi_label_annotate(
    ['assets/altera.jpg', 'assets/chibi_gil.jpg','assets/chibi_saber.jpg'],
    options={'cute':['yes','no'], 'saber':['yes','no'],'colors':['blue','gold','white','red']},
    display_fn=lambda filename: display(Image(filename))
    )

Preview:

New in v0.3.0: colors!

Additional Notes

• skip readded the old skip button for clarity. In the case you dont want to label any annotations for a sample you can press skip or done but having skip is probably more clear.

I added some additional buttons as well to the multi-label script. Since we want to be able to select more than one category I added a few extra buttons.

• done to press after all relevant fields have been clicked.

• back button in case you want to go back to a previous image. WARNING this will erase that item from the dictionary so you will have to click through all the classes you want to mark for that previous image again.

• clear current WARNING will delete the current image from the dictionary on the back end so you can re enter the classes in case you messed one up.