[QUESTION] Customising "category_id" values when exporting a COCO-formatted dataset

[astringfield]

I have a custom COCO-formatted dataset loaded into FiftyOne and I'm trying to export the images and annotations in COCO format:

import fiftyone as fo

custom_coco_dataset_view.export(
        export_dir=str(output_path),
        dataset_type=fo.types.COCODetectionDataset,
        label_field="segmentations",
    )

Currently, the COCO category_id values in the export COCO JSON file start at 0.

Is it possible to modify this behaviour so that the category_id values start at 1?

Examples below of the current export results and the export results I'm hoping to see.

Current export format (categories start at 0)

"categories": [
    {
        "id": 0,
        "name": "apple",
        "supercategory": ""
    },
    {
        "id": 1,
        "name": "orange",
        "supercategory": ""
    },
    {
        "id": 2,
        "name": "banana",
        "supercategory": ""
    },
]

Desired export format (categories start at 1)

"categories": [
    {
        "id": 1,
        "name": "apple",
        "supercategory": ""
    },
    {
        "id": 2,
        "name": "orange",
        "supercategory": ""
    },
    {
        "id": 3,
        "name": "banana",
        "supercategory": ""
    },
]

Appreciate any help - thanks!

[benjaminpkane]

HI @astringfield. I do not believe this is supported by default. One option is subclass COCODetectionDatasetExporter and override the _parse_classes method. It looks like that is where the class mapping is initialized. (Note the safer solution is to re-implement the entire setup method as that is a public contract)

Contributions are also welcome if you see an opportunity to add category_id configuration out of the box!

[brimoor]

#4354 adds support for this!

categories = [
    {'supercategory': 'person', 'id': 1, 'name': 'person'},
    {'supercategory': 'vehicle', 'id': 2, 'name': 'bicycle'},
    ...
    {'supercategory': 'indoor', 'id': 90, 'name': 'toothbrush'}
]

dataset.export(
    export_dir="/path/for/export",
    dataset_type=fo.types.COCODetectionDataset,
    categories=categories,
)