Bounding box annotation

[thomassajot]

A useful feature for object detection is to define the bounding box (or even polygon) around one or many objects in an image.

[thomassajot]

Keeping the prediction made by the model on the remaining images would speed up the labelling process.
This UI is build using jupyter-innotater

[janfreyberg]

Hi,

thanks for raising this! I am actually currently working on this. I'm developing the bounding box / polygon interface here: https://ipyannotations.readthedocs.io/en/latest/

For the integration, really the only thing missing is to document it. If you want to try it, you can do:

pip install superintendent ipyannotations

then run this in a notebook:

from ipyannotations import PolygonAnnotator
from superintendent.base import Labeller

input_widget = PolygonAnnotator(classes=["pigeon"])

labeller = Labeller(
    features=["img/vdnkh.jpg"],  # your list of images
    input_widget=input_widget,
    display_func=input_widget.display
)

labeller

[thomassajot]

Hi,

This is great !
I believe the classes argument of PolygonAnnotator should be called options.

Unfortunately I did not manage to display the image.

from ipyannotations import PolygonAnnotator
from superintendent.base import Labeller

input_widget = PolygonAnnotator(options=["pigeon"])

labeller = Labeller(
    features=["my_image.jpg"],
    input_widget=input_widget,
    display_func=input_widget.display
)

labeller

Would you know about this issue ?

[janfreyberg]

Yes, you're right, that's a mistake on my side.

I'm not sure why the image is not displaying... is the image stored in the same directory as the notebook you're working in?

And if you have an older version of jupyter notebook installed, you may have to also run this command:
jupyter nbextension enable --py --sys-prefix ipycanvas

(see here )

[thomassajot]

Installing the extension solves the display issue.
Thanks !

[thomassajot]

There is a subsequent bug.
Running the code once creates a labeller.new_labels with one element in the list [None]
Running the code a second time, labeller.new_labels is now a list of two elements [None, None]

[janfreyberg]

OK, thanks for letting me know - I'll take a look

[janfreyberg]

I've figured it out, it's a silly bug on my part. I'm pushing a fix now, plus some documentation of this example!

[janfreyberg]

This bug is now fixed (see https://github.com/janfreyberg/superintendent/releases/tag/0.5.1) - make sure to upgrade to the latest version from PyPI. As part of this release I also added documentation for using ipyannotations widgets: https://superintendent.readthedocs.io/en/latest/modularity.html

[thomassajot]

Great ! Thank you for the fix.