Synthetic data to common format, e.g., COCO

Synthetic data creation works great! I want to use this for training my ML models of course.

Are there any known/existing/working converters from the Isaac sim ‘numpy’ format that can result in converting to a more common format? Like coco. (The built-in kitti writer is highly limited.) I need class and instance segmentation.

My options, other than an existing converter are to write my own converter, or perhaps customize the import in ‘fiftyone’ - Loading Datasets From Disk — FiftyOne 0.16.6 documentation . (That would seem the best, given their rich translation capabilities…)



Hi Peter,

glad to hear that you find the synthetic data generation tool useful for your project.
Here you can find more information about OV Replicator and specifically how to write a custom writer: Custom Writer — Omniverse Extensions documentation