How would you decide whether to use a CNN, transformer, or another architecture for a perception task in Autonomous Solutions?
