Closed mi2celis closed 2 years ago
The COCO format describes a full dataset, whereas Pascal VOC describes annotations for an image. Therefore, something like image.to_coco
wouldn't work.
Check the fathomnet-generate source to get an idea of how the --format coco
flag works. I wrote a small library (coco-lib) with COCO data models as Python dataclasses to facilitate this process.
I am able to generate COCO style json files by using fathomnet-generate with the option --format coco.
In your Colab tutorial you explain how to use image.to_pascal_voc to generate Pascal_VOC dataset. Is there an equivalent way to generate a COCO dataset?