image_bbox_app

This is an app that utilizes Gradio to create object-level annotations for images.

Install

conda create -n image_app python=3.9

conda activate image_app

pip install -r requirements.txt

How to run

Assuming you have a list of images within your image folder, create a txt file containing all the object classes similar to example_classes.txt

Then:

python image_annotation_app.py --input_image_directory image --input_image_classes_path example_classes.txt --output_annotation_json image_annotations.json

This will create a local and sharable url that can be used to annotate any images within the folder.

How to annotate

1. Select an the image you want to annotate.
   1. You can use the previous image & next image to go to the previous / next image within the folder
   2. Alternatively, you can type the name of the image you want within the Image textbox and press Go to Image
   3. You can also use the slider to jump to an image
2. Use the slider for (x_min, y_min, x_max, y_max) to create a bounding box for that image. These values can also be modified if you write within the upper right textbox of each slider panel.
3. Use the dropdown options to select a class for the above bounding box.
4. Click add bbox annotation. This will save the annotation within the output_annotation_json path. If you go back to this image the annotation will be loaded and shown in the Image annotation metadata panel.

How to remove an annotation

At the moment you can remove the latest annotation for an image by just clicking the Remove bbox annotation button