Object detection pretrained model and domain randomisation


With the help of Object Detection Pipeline documentation I was able to run inference using the pretrained model.

It is also to my understanding that Isaac Sim has a built-in domain randomisation feature for training.

However, since it wasn’t clearly stated, is the pretrained model trained with or without the domain randomisation?



Without Domain Randomization.


so is there a guide on how to train for a new object using domain randomization?


Hi dmtesfav,

You can look at the following links for the guide:

  1. https://docs.nvidia.com/isaac/isaac/packages/ml/ml.html
  2. https://docs.nvidia.com/isaac/isaac/doc/simulation/unity3d.html#randomization
  3. https://docs.nvidia.com/isaac/isaac/packages/detect_net/doc/detect_net.html?highlight=randomization

These links will give you an idea of how to train using domain randomization.