Regarding selection criteria of subset of database images & graph connectivity

[HyeonJaeGil]

Hi, thanks for your great work.

After reading your paper, a few questions arised.

1. When using smaller database images (K=1,2,5), how did you choose the subset? Is there any crieteria, or if random, isn't the result varies based on the seed?
2. When extracting geometric features, may I understand that the graph (whose nodes are objects) is fully-connected and each edge contributes to the geometric features? I was asking this because the figures in your paper (fig1, for example) only includes fewer edges than I thought.

Thanks in advance.
Best,
Hyeonjae