Scrfd Computation distribution

Scrfdd

What is meant by computation distribution in paper is it model parallelism. IF so, then you are picking 320 model based on your Manuel adjustments and Manuel population Creation, and checking their computation distribution. Then models, and taking model with highest AP in neck head and backbone all. Am I correct? Also I am guessing in Nas here there is no evolutionary algorithm or reinforcement learning here. Moreover you are using resnet with blocks adjusted according to regent. Sorry it is long.