Ever since I learned about SLAM/SfM I was wondering why are both feature based and direct methods outputting point clouds. If you detect a line of points why st

Ever since I learned about SLAM/SfM I was wondering why are both feature based and direct methods outputting point clouds. If you detect a line of points why store it as points when you could store a vector representation, line primitive. A win in storage, tracking and potentially drift correction (easier to correlate?).