Synthetic data generation using replicator

I have gone through the replicator tutorials and tried out the following demos

  1. Replicator composer using a yaml
  2. Offline dataset generation using a .py script

What is a general rule of thumb in picking one of these methods to generate synthetic data? I have found the first method to be helpful since I can do a lot of customization even in a headless mode with the only downside being that my annotations aren’t saved in KITTI format. On the contrary, the second option isn’t as customizable but saves the annotations in the KITTI format.
Any help will be greatly appreciated.

It seems that for now the second option is the officially supported one and composer will be deprecated. Check out my post and the answer I got here Replicator - nested loops or task queque This might help you out.