search

yblin / global_l0

Global L0 algorithm for regularity-constrained plane fitting
52 stars 14 forks source link

the algorithm doesn't converge #2

Closed h1063135843 closed 3 years ago

h1063135843 commented 3 years ago

The program works well in some situation. But when I attempt to detect planes in some bigger pointclouds, the program seems to be stuck in a endless loop and I can't have a result. Does the algorithm always converge?

h1063135843 commented 3 years ago

image I print the subset size after each iteration, and the size doesn't decrease to meet the breaking condition. On the contrary it always increase.

yblin commented 3 years ago

Strange problem, can you send me the data

发件人：duan-she-li notifications@github.com 发送日期：2020-12-03 19:29:36 收件人：yblin/global_l0 global_l0@noreply.github.com 抄送人：Subscribed subscribed@noreply.github.com 主题：Re: [yblin/global_l0] the algorithm doesn't converge (#2)

I print the subset size after each iteration, and the size doesn't decrease to meet the breaking condition. On the contrary it always increase. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

h1063135843 commented 3 years ago

Strange problem, can you send me the data 发件人：duan-she-li notifications@github.com 发送日期：2020-12-03 19:29:36 收件人：yblin/global_l0 global_l0@noreply.github.com 抄送人：Subscribed subscribed@noreply.github.com 主题：Re: [yblin/global_l0] the algorithm doesn't converge (#2) I print the subset size after each iteration, and the size doesn't decrease to meet the breaking condition. On the contrary it always increase. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

Thanks for your reply. Finally the program works well, and it is just too slow in my case(18 hours). The subset size always increases firstly. Increasing the number of min support points will accelerate the process for big pointcloud. Furthermore, computing distance2 in the metric function also cost too much time. I am curious about it whether it is the same as the rmsn metric mentioned in your article.

yblin commented 3 years ago

Thanks for your reply. Finally the program works well, and it is just too slow in my case(18 hours). The subset size always increases firstly. Increasing the number of min support points will accelerate the process for big pointcloud. Furthermore, computing distance2 in the metric function also cost too much time. I am curious about it whether it is the same as the rmsn metric mentioned in your article.

Do you mean RMSE? I think it is the same