Sub-graph implementation

[chenming-wu]

Hi authors thanks for the great work and sharing the code! I am confused about the sub-graph mentioned in the paper while the code seems like using a complete graph instead of a 4-neighbor subgraph? Is that because each fragment contains 4 views so there is no need to use 4-neighbor subgraph or for another reason? Thanks.

[chenming-wu]

BTW, it looks like the implementation in the code first reconstructs a global alignment for keyframes and then registers the frames of each fragment to the keyframe, which is also aligned with the global point cloud. Is my understanding correct?