Point Cloud Generation - Very Slow after 74% with Low CPU

[RoryG]

Hi, This is not really a problem but I am curious to know what's going on. I am working with a 9,210 20MP image dataset. Currently generating the point cloud, the depth maps took about six hours to build, it is now generating the point cloud but at 74% it slowed down with only about 5% CPU utilisation. The total time currently is 22 hours 21 minutes and it is estimating around four hours to complete, although it has taken four hours to go from five hours remaining to four hours remaining! Workstation spec is an Intel i9 10980XE 18 core with 256GB RAM and 2 x GF RTX 3070 Ti. Images are on a fast NVMe SSD.

The images are from two drones, a Matrice 4E and a Mavic 3E. Alignment was done on high, and I didn't use the fit additional corrections option when optimising cameras which can slow things down. Point cloud generation was set to high with aggressive filtering, and both calculate point colours and point confidence enabled.

I'm curious to know why it reaches a point where there is little CPU use? Is there anything I can do to speed things up in the future? I regularly process datasets of around 3,000 to 5,000 images and it seems to use a lot more CPU to almost the end of point cloud generation.

TIA Rory

[olihar]

Yes this step is single threaded.

You could try the newest test version, I think there are some improvements there.

[RoryG]

Yes this step is single threaded.

You could try the newest test version, I think there are some improvements there.

Ah, that makes sense, thanks for the info. It completed after 26 hours so all good. I’ll maybe try the pre release version on the next survey.

Cheers,

Rory