ComicAnnotator

An annotation tool for comics. Written in Tkinter, configured for panel, face, body, speech bubble, and narrative annotation.

Features

• Annotate images with face, body, panel, speech bubble, narrative, and speech bubble tail bounding boxes.
• Shift and resize the drawn boxes with the edit mode.
• Add associations between faces, bodies, speech bubbles, and speech bubble tails.
• With identification mode, select bodies that belong to the same person (a circle appears with the same color in the center of body box).
• Undo changes if something is done wrong.

Folder Structure

• Add your images to annot_images folder.
• Inside meta_data folder:
  • Place your list of image files inside the files_list.txt file.
  • Create a json file named users_range.json and define your specific conditions there (which user will see which range of images):

{
    "test": [
        0,
        2
    ],
    "user1": [
        0,
        20
    ],
    ...
}

• Create a results directory, where the annotation results are written. The annotation file name will be results/{img_name}.png and outputs will be in this format:

### BOXES
index,x1,y1,x2,y2,obj_name
...

### ASSOCIATIONS
index1,index2,index3,...
index4,index5,...

### IDENTIFICATIONS
index1,index2,index3,...
index4,index5,...
...

### TEXTS
box_id1,text1
box_id2,text2
...

Usage

• Select the object to annotate,
• Long press and drag for making a bounding box,
• After finishing all the objects, press to "a",
• Associate each object by clicking on their boxes,
• When one association is finished, click to "n" for starting a new association.
• Press "e" to edit your boxes.

Screenshot

• General View of the Application

• View During the Text Transcription Mode

TODO

• Add gaze direction labeling
• Embed a detector model to help with the box annotation