On the choice of the number of pretrained models

[charmeleonz]

Hi, I'd like to ask how is the value 1000 determined as the number of pretrained ResNet18 models, given that this means some significant training and storage costs. Also, how will lower values impact the performance? Many thanks.