colmap sparse reconstruction take a lot of time

2,769 views
Skip to first unread message

change1...@gmail.com

unread,
Oct 25, 2018, 8:44:58 PM10/25/18
to COLMAP
  • I used colmap for sparse reconstruction,and the number of images was about 5,000.In the colmap source paper,5,000 images take about 1,000 seconds,but it took me nearly 10 days to get the results.I use release mode and multi-thread to run it.I really don't kown why there is a such speed difference.Has anybody done the same experiment?

Doan Anh Dung

unread,
Oct 26, 2018, 2:19:00 AM10/26/18
to col...@googlegroups.com
You can use PBA for faster sparse reconstruction, but camera model has to be restricted to SIMPLE_RADIAL.

On Fri, Oct 26, 2018 at 11:15 AM <change1...@gmail.com> wrote:
  • I used colmap for sparse reconstruction,and the number of images was about 5,000.In the colmap source paper,5,000 images take about 1,000 seconds,but it took me nearly 10 days to get the results.I use release mode and multi-thread to run it.I really don't kown why there is a such speed difference.Has anybody done the same experiment?

--
You received this message because you are subscribed to the Google Groups "COLMAP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to colmap+un...@googlegroups.com.
To post to this group, send email to col...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/colmap/0ded14bf-106c-497a-810b-9a0769d594fd%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

change1...@gmail.com

unread,
Oct 27, 2018, 11:21:39 PM10/27/18
to COLMAP
Thank you very much for answering my question.By using PBA, do you mean multi-thread BA or GPU BA.I have restricted the camera model to SIMPLE_RADIAL,and it seems that colmap can only use CPU for BA and cannot use GPU.When I use multi-thread to reconstruct, It still takes a long time to reconstruct, it takes about 4 hours to reconstruct a model contains only 400 images. The server CPU I used is ok, and I set the number of threads to 32, so maybe it's not a machine problem.I really don't know why there's such a big difference in speed

在 2018年10月26日星期五 UTC+8下午2:19:00,Anh-Dzung Doan写道:
You can use PBA for faster sparse reconstruction, but camera model has to be restricted to SIMPLE_RADIAL.

On Fri, Oct 26, 2018 at 11:15 AM <change1...@gmail.com> wrote:
  • I used colmap for sparse reconstruction,and the number of images was about 5,000.In the colmap source paper,5,000 images take about 1,000 seconds,but it took me nearly 10 days to get the results.I use release mode and multi-thread to run it.I really don't kown why there is a such speed difference.Has anybody done the same experiment?

--
You received this message because you are subscribed to the Google Groups "COLMAP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to colmap+unsubscribe@googlegroups.com.

Doan Anh Dung

unread,
Oct 28, 2018, 1:46:35 AM10/28/18
to col...@googlegroups.com
Yes, PBA is for multi-core CPU, I think COLMAP hasn't supported sparse reconstruction on GPU

On Sun, Oct 28, 2018 at 1:51 PM <change1...@gmail.com> wrote:
Thank you very much for answering my question.By using PBA, do you mean multi-thread BA or GPU BA.I have restricted the camera model to SIMPLE_RADIAL,and it seems that colmap can only use CPU for BA and cannot use GPU.When I use multi-thread to reconstruct, It still takes a long time to reconstruct, it takes about 4 hours to reconstruct a model contains only 400 images. The server CPU I used is ok, and I set the number of threads to 32, so maybe it's not a machine problem.I really don't know why there's such a big difference in speed

在 2018年10月26日星期五 UTC+8下午2:19:00,Anh-Dzung Doan写道:
You can use PBA for faster sparse reconstruction, but camera model has to be restricted to SIMPLE_RADIAL.

On Fri, Oct 26, 2018 at 11:15 AM <change1...@gmail.com> wrote:
  • I used colmap for sparse reconstruction,and the number of images was about 5,000.In the colmap source paper,5,000 images take about 1,000 seconds,but it took me nearly 10 days to get the results.I use release mode and multi-thread to run it.I really don't kown why there is a such speed difference.Has anybody done the same experiment?

--
You received this message because you are subscribed to the Google Groups "COLMAP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to colmap+un...@googlegroups.com.

To post to this group, send email to col...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/colmap/0ded14bf-106c-497a-810b-9a0769d594fd%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "COLMAP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to colmap+un...@googlegroups.com.

To post to this group, send email to col...@googlegroups.com.

change1...@gmail.com

unread,
Oct 28, 2018, 9:09:06 PM10/28/18
to COLMAP
Thank you! I would like to ask whether the speed of colmap in your experiment can reach the speed in the paper?

在 2018年10月28日星期日 UTC+8下午1:46:35,Anh-Dzung Doan写道:

Doan Anh Dung

unread,
Oct 28, 2018, 9:22:17 PM10/28/18
to col...@googlegroups.com
I think the speed of sparse reconstruction also depends on the number of matching pairs. I have read through the paper, authors only consider 100 nearest neighbors per each image. If you use exhausted matching, the number of matching pairs would be very large, so BA is slow. It also depends on number of CPU cores they used, but it have not been mentioned in the papers :)


change1...@gmail.com

unread,
Oct 29, 2018, 2:37:00 AM10/29/18
to COLMAP
thank you very much.

在 2018年10月29日星期一 UTC+8上午9:22:17,Anh-Dzung Doan写道:
Reply all
Reply to author
Forward
0 new messages