Why is the number of blocks per group different in Imagenet?

[Rahim16]

In contrast to cifar code, for imagenet the number of blocks in groups is not calculated automatically, but provided manually, where the third group gets the most residual blocks as shown in the code snippet below. What is the reason behind this distribution of blocks? Have you also tried to create the same number of blocks per group as done for CIFAR datasets?

local cfg = {
         [18]  = {{2, 2, 2, 2}, 512*width, basicblock}, -- lea  as is
         [34]  = {{3, 4, 6, 3}, 512*width, basicblock}, -- leave as is
         [50]  = {{3, 4, 6, 3}, 512*bottle, bottleneck},
         [101] = {{3, 4, 23, 3}, 512*bottle, bottleneck},
         [152] = {{3, 8, 36, 3}, 512*bottle, bottleneck},
      }