For machine vision problems, synthetic data with domain randomisation definitely works, provided, that is, you are able to generate a large enough number of different modes of variation.
In fact, it works so well it feels like cheating.
In fact, it works so well it feels like cheating.
No comments yet.