Questions about PnP solutions

danini / graph-cut-ransac

The Graph-Cut RANSAC algorithm proposed in paper: Daniel Barath and Jiri Matas; Graph-Cut RANSAC, Conference on Computer Vision and Pattern Recognition, 2018. It is available at http://openaccess.thecvf.com/content_cvpr_2018/papers/Barath_Graph-Cut_RANSAC_CVPR_2018_paper.pdf

Other

426 stars 92 forks source link

Questions about PnP solutions #29

Open uurbsrn4d opened 3 years ago

uurbsrn4d commented 3 years ago

Hi @danini

I was trying your PNP/Ransac implementation for python. I had a few questions.

It seems like opencv version ~50x faster. Is this expected because of the different pnp and ransac solutions used in methods or something is wrong ? By the way Pose solutions are nearly same for both versions.

My second question is are you planing to add other pnp methods such as SQPNP?

Thanks for your great work and sharing.

danini commented 2 years ago

Ah sorry for not answering earlier. This definitely seems some problem, I will investigate it tomorrow. I am planning on adding yes some other PnP solver since the current one, DLSPnP is not too reliable in my tests.

uurbsrn4d commented 2 years ago

Is there any update on this issue?

danini commented 2 years ago

Actually, I have slightly more time than usual so I will look into this tomorrow and will, hopefully, solve it.

danini commented 2 years ago

I replaced all the solvers. It know works better than OpenCV on the example I tried. https://github.com/danini/graph-cut-ransac/blob/master/examples/example_absolute_pose.ipynb

I still need to update the example since I added outliers artificially, but the algorithm should work and be fast now. Let me know if there still is a problem.