Looking hg-s2-b1

hi there
in your comparision graph you are comparing stack 8 vs stack 2 . Are there trained weights available for stack 2 ?
2) How much GPU do you need with batch size of 8 to 10 for stack 2 and stack 8 .

3) which is better hg_gn or hg
4) is HG built using resnet?
