workflow

[pixel_Killer]

Hello,

I've been using Agisoft for a while now for personal projects and I thought I share some of my findings.
I don't have a super computer so I've always needed to find a way to optimize my projects... My machine is an 8core amd, gtx 660 2gb, with 16 gb of ram (hoping to upgrade to 32). So when processing my projects I usually run out of ram.

I shoot with a 5D mark II and my projects image count run bw 25-50, depending on my subject. Most of the time I can process my project at medium resolution without my computer locking up. So obviously when changing the setting to ultra, my machine will always lockup.

My main issue was losing precious calculation time when I ran into a memory block. I wanted a two step process.
1. calculate depth map and save.
2. build geo without calculating depthmap

So my solution was to export a dense pointcloud, choose ultra settings, which saves a file to text and use far less ram. Upon finishing I would save since now I have my depth maps calculated at ultra high settings.

I will now build geo, with build depth map unchecked, and for my personal needs I limited my mesh to 5million. This method so far has saved me a lot of time compared to doing everything in one go.

Hope this may be helpful to somebody.

I do have a question about depth map, if anyone can chime in. I've notice sometimes that there will be an x amount of image where agisoft could not calculate a depth map and I don't know why that is. Can anyone explain this? I try to shoot most of my images with a greater depth of field and decent amount of overlap. Is agisoft expecting a sequence of images to calculate depth maps? If I break the sequence and shoot another angle, does that hurt the depth map calculation process?

[Infinite]

Really useful post. Interesting idea for process and saving time.

I also notice that some depth maps don't build and wondered why. Not always but sometimes 1-5 may not.

[FoodMan]

I think it is because after you optimize the alignment and delete some points, some images may have lost too much points...? so maybe Pscan can't use them...

well at least that is what I noticed..

[RalfH]

Nice workflow, and very useful as I am working under similar RAM constraints.

As for the depth maps, I have been wondering as well. All cameras are aligned, no changes were made to the point cloud or the camera alignment, but then not all depth maps are built.

[Alexey Pasumansky]

Hello pixel_Killer,

Actually you can build depth maps using corresponding command from Photos pane context menu (with all cameras selected) in case you do not need Dense point cloud export, but require only depth maps to be calculated.

And about missing depth maps: there are two main reasons for that. The first one, it is outside the bounding box, and the second, the camera alignment is unreliable.

[RalfH]

If PhotoScan has a way of deciding whether or not a camera alignment is reliable, is it possible to output this somehow? For example, output a "reliability index" or something like that as a number attached to the depth maps? This could help the user in deciding where there may be a problem in the project.

[Alexey Pasumansky]

Hello RalfH,

It is mostly based on the number of valid matches.
View matches window is accessible from Tools menu.