Pointrend model - Tracing or Scripting

[amal-r1]

Does it make sense for us to do tracing to convert the pointrend model to Torchscript?
I am a bit confused here, as pointrend is a dynamic model as per my knowledge(extremly sorry if I am wrong) and tracing won't be able to capture the dynamic elements of the model, so will its performance be similar?
Alternatively, is scripting the only approach that can effectively capture the complexities of the PointRend model and ensure similar performance to the original PyTorch model?.