search

LimHyungTae / patchwork

SOTA fast and robust ground segmentation using 3D point cloud (accepted in RA-L'21 w/ IROS'21)
MIT License
487 stars 74 forks source link

the performance of the proposed components #8

Closed shouldnotfail closed 2 years ago

shouldnotfail commented 2 years ago

Thanks for sharing the awsome works pathwork and erasor! And I wonder can CZM & GLE lead to performance gain when using erasor?

LimHyungTae commented 2 years ago

Sure I guess, it would reject some false positive patches!

shouldnotfail commented 2 years ago

Aftering reading the code carefully, I found that the ground likelihood is represented by some discrete states, which is not same as the description in paper. The reason is the difficulty of parameters tuning?

LimHyungTae commented 2 years ago

Above all things, thank you for interest 😍! In fact, the GLE is the equation seems like fancy but eventually it means binary decision haha....

LimHyungTae commented 2 years ago

I'll share my slides in IROS'21 to help you understand better, which below pipeline is identical with the GLE equation step-by-step! image image image

shouldnotfail commented 2 years ago

got the same feeling! the invese sensor model in lidar based freespace detection always turn to be binary classifier too. Thanks again for the reply!

LimHyungTae commented 2 years ago

Cheers~ Hope your search improves my method!