Visualizer should demonstrate source<->target point-cloud correspondences, similar to the example code.
Correspondence strength determined by MaskNet should be represented as color map of the connecting lines.
However, for source we should use the current position of the points in the source point cloud (at the current iteration), and the target can remain the same -- the target point cloud.
This will help us analyze whether the 3 Gauss-Newton iterations are really enough to achieve proper convergence for every time range.
Algomorph
commented
2 years ago
Visualizer should demonstrate source<->target point-cloud correspondences, similar to the example code.
Correspondence strength determined by MaskNet should be represented as color map of the connecting lines.
However, for source we should use the current position of the points in the source point cloud (at the current iteration), and the target can remain the same -- the target point cloud.
This will help us analyze whether the 3 Gauss-Newton iterations are really enough to achieve proper convergence for every time range.