search

prs-eth / OverlapPredator

[CVPR 2021, Oral] PREDATOR: Registration of 3D Point Clouds with Low Overlap.
https://shengyuh.github.io/predator/index.html
MIT License
512 stars 73 forks source link

How to compute overlap ratio #7

Closed Gilgamesh666666 closed 3 years ago

Gilgamesh666666 commented 3 years ago

Hi, @ShengyuH

Thank you for sharing your excellent work, I try to calculate the overlap ratio of the 3DMatch dataset.I follow the code of 3DMatch to calculate the overlap ratio of the datasets provided by here, but the result seems be different from the gt_overlap.log. So i wonder if you could give me some hint about the details.Thank you very much!

ShengyuH commented 3 years ago

hi,

Thanks for your interest in our work. I've now commited our script used to compute overlap ratio.

First, we are providing subsampled point clouds, to compute overlap ratio, you should actually use the raw point clouds which are way denser.

Second, as we've mentioned in the main paper, there's a bug in the script provided by 3DMatch team. Let's say we have M and N points for source point cloud P and target point cloud Q, for every point p_i in P, we do NN search and retrieve the nearest point in Q and if the distance is within a threshold, then it's counted as in the overlap region. We end up with m points in the overlap region, then the overlap ratio is m/M. However, in the script from 3DMatch team, they are actually using m/N. If you carefully check the overlap ratios you obtained with their script, you can actually find some values above 1 which is not reasonable.

Please let me know if this helps.

Best, Shengyu

Gilgamesh666666 commented 3 years ago

Thank you for your elaborated reply! I get the problem.

Best, zebai