Score: 0 / 4
. What does triangulation compute, given two camera projection matrices and a matched 2D point pair?
Triangulation intersects the two rays (one per camera) that could have produced the matched 2D observations, recovering the 3D point via DLT/SVD.
. What is the purpose of the cheirality check (requiring positive depth in both cameras) after triangulation?
The epipolar constraint doesn't rule out points that triangulate behind a camera; requiring positive depth in both views is a free extra filter that catches such bad matches.
. Why does the reconstruction from *estimated* poses need to be multiplied by the true baseline length before it matches the true-pose reconstruction?
Since t from recoverPose is unit-length, every triangulated point comes out a fixed factor smaller than reality; rescaling by the true baseline fixes the metric scale (using outside knowledge, since two views alone can't supply it).
. What does bundle adjustment do that distinguishes a real multi-view SfM system from the simple two-view pipeline in this lesson?
As more views are added, per-view pose errors compound; bundle adjustment jointly refines every camera pose and 3D point at once to minimize the total reprojection error, correcting for that compounding.