Generate annotations data in coco.

AnuragTimilsina · February 4, 2023, 1:38pm

Is there a direct option to export annotated polygon image data in the coco format? If not, what's the best way to convert the default prodigy .jsonl file to the coco format?

koaning · February 6, 2023, 10:03am

Have you seen this answer?

It might be a nice example to work from. Prodigy currently does not offer an automated way to get data into the COCO format, but there might be some community examples out there that do.

ryanwesslen · April 6, 2023, 1:45pm

hi @AnuragTimilsina,

I created a simple script that converts .jsonl to coco:

gist.github.com

https://gist.github.com/wesslen/f345e7155f2ce7800e52bdda66a2c9b2

jsonl-to-coco.py

import json
from typing import List
import srsly
import typer

app = typer.Typer()


def convert_to_coco(input_file: str, output_file: str):
    # Load the JSONL file using srsly

This file has been truncated. show original

prodigy-boundingbox-input.jsonl

{"image":"https://images.unsplash.com/photo-1434993568367-36f24aa04d2f?w=400","width":400,"height":267,"spans":[{"label":"SKATEBOARD","color":"yellow","x":47.5,"y":171.4,"width":109.1,"height":67.4,"points":[[47.5,171.4],[47.5,238.8],[156.6,238.8],[156.6,171.4]],"center":[102.05,205.1],"type":"rect"},{"label":"PERSON","color":"cyan","points":[[256,39.5],[237,78.5],[269,116.5],[286,67.5]],"type":"polygon"}]}
{"image":"https://images.unsplash.com/photo-1434993568367-36f24aa04d2f?w=500","width":500,"height":267,"spans":[{"label":"SKATEBOARD","color":"yellow","x":47.5,"y":171.4,"width":109.1,"height":67.4,"points":[[47.5,171.4],[47.5,238.8],[156.6,238.8],[156.6,171.4]],"center":[102.05,205.1],"type":"rect"},{"label":"PERSON","color":"cyan","points":[[256,39.5],[237,78.5],[269,116.5],[286,67.5]],"type":"polygon"}]}

prodigy-coco.json

{"info": {}, "licenses": [], "categories": [{"id": 1, "name": "SKATEBOARD", "supercategory": ""}, {"id": 2, "name": "PERSON", "supercategory": ""}], "images": [{"id": 1, "width": 400, "height": 267, "file_name": "https://images.unsplash.com/photo-1434993568367-36f24aa04d2f?w=400"}, {"id": 2, "width": 500, "height": 267, "file_name": "https://images.unsplash.com/photo-1434993568367-36f24aa04d2f?w=500"}], "annotations": [{"id": 1, "image_id": 1, "category_id": 1, "iscrowd": 0, "bbox": [47.5, 171.4, 109.1, 67.4], "area": 7353.34, "segmentation": [[47.5, 171.4, 47.5, 238.8, 156.6, 238.8, 156.6, 171.4]]}, {"id": 2, "image_id": 1, "category_id": 2, "iscrowd": 0, "segmentation": [[[256, 39.5], [237, 78.5], [269, 116.5], [286, 67.5]]]}, {"id": 3, "image_id": 2, "category_id": 1, "iscrowd": 0, "bbox": [47.5, 171.4, 109.1, 67.4], "area": 7353.34, "segmentation": [[47.5, 171.4, 47.5, 238.8, 156.6, 238.8, 156.6, 171.4]]}, {"id": 4, "image_id": 2, "category_id": 2, "iscrowd": 0, "segmentation": [[[256, 39.5], [237, 78.5], [269, 116.5], [286, 67.5]]]}]}

It can be run by:

python jsonl-to-coco.py prodigy-boundingbox-input.jsonl prodigy-coco.json

I haven't tested it for all cases, but it may help you or someone else.

If anyone finds issues or has suggestions, let us know.