Messages

31

Bug Reports / Re: Memory leak in point matching

« on: November 22, 2013, 03:09:59 PM »

Also, If I have a partial list of camera coordinates (for example those generated in a lower res run for those parts of the model which are constructed ok) how will photoscan treat this if I select ground control preselection?

In other words do you need a complete list of camera positions in order to use ground control preselection? And if not, how does it treat cameras without location?

Thanks.

32

Bug Reports / Re: Memory leak in point matching

« on: November 22, 2013, 02:52:30 PM »

Hi Alexey,

Unfortunately I don't have any camera coordinates, but I have tried using the generic pair selection ( I presume this uses an initial low resolution matching to select the pairs?). Sadly though, my data set is pretty poorly textured, so the generic pair selection is insufficient to align my photos. Hence I chose to bite the bullet and wait for the much longer processing for matching all pairs. Sadly though, this crashes for runs longer than about 2.5 days - even with such a large dataset , it really shouldn't need more than 300GB of RAM!

How does logging work in the linux version of agisoft? Does it write a log file automatically? If so, where to?

Many thanks.

33

Bug Reports / Re: Memory leak in point matching

« on: November 21, 2013, 06:37:25 PM »

I'm using a Mandriva Linux system on a server, running agisoft remotely via ssh.

The photos are 3648 x 2736px, high accuracy, with no pair pre-selection. Anything above around 15000 point limit results in run times so long that the memory leak will fill all 384GB of RAM.

Where is the logfile written to on the linux based version?

Thanks

EDIT:
I have been using a few different numbers of photos, between about 1700-2250 photos.
The photos are pretty poorly textured, hence running at a large number of points.

34

Bug Reports / Memory leak in point matching

« on: November 21, 2013, 12:28:44 PM »

I am using a 16core system with 384GB RAM - Running out of RAM should not be an issue!

When I run a large photoset (1500-2500 photos) I expect the matching to run for somewhere between 5-8 days, but there appears to be a slow memory leak. As a result, once the matching starts, the amount of memory in use slowly increases without halt, until all 384GB is in use... but it doesn't crash immediately, it only crashes once the matching is basically finished!

I have tried this on both 0.9.1 and 1.0.0. The memory appears to leak more slowly on the beta version 1.0.0, but it is still a problem for anything longer than about 2 days.

Any help much appreciated.