Tools for training and managing MS COCO datasets (work-in-progress).
You can use annotate.py to create annotations from a jpg.
Let's say you have an image of a comic that looks like this, and you want to create annotations of the speech bubbles.
Go to any image editing tool and create a jpg, using any color where the speech bubbles are, and black for the portions we want to leave out.
Save the jpg in the annotations directory of your dataset, then run annotate.py to get the mask in coco's polygon format for segmentation and save it in a json file.
You can use the show_annotations flag to draw the polygons for testing/debugging, which will render an image that looks like this.
