PhotoScanPro - not responding

[mwittnebel]

Hello all,

I have created a geometry with ~200million faces.
If i now click on file, in the menu bar, the Software (PhotoScanPro) is not responding.
Is this problem known ?

(Is this maybe a problem of my GPU ?)

CPU: i7-3930K
RAM: 32 GB
GPU: NVIDIA 630 2GB

Regards Marcus

[Wishgranter]

its overkill on your PC. not even my dual xeon could handle that.... main limit is there GPU ram - if it not fit in GPU ram it will use main memory, and that is very slow because of how it works...... for 2 GB GPU ram wold not exceed 20-25 mil trias..... not counting texture size...... a 8k textr is 192 MB, 16k is 768 MB vRam....

[mwittnebel]

i have loaded a file with ~43million faces and the GPU ram usage is increased by ~ 13 % (270MB).

• align done
• geometry done
• texture not done

Is the number of faces linear to the needed GPU ram ?

Is it possible to reduce the displayed faces in PhotoScan, without reduce the amount of faces in the file ? (more or less, like a WMS)

[Wishgranter]

what sort of geometry you do ?? how you process then so big files ?? edit, cleaning stuff and etc ??

No its not possible. but it could be done with help of BATCH process, so you export H U G E OBJ file with all 200+ mil trias. then decimate in BATCH and see say 50 mil results......

[mwittnebel]

Processing Information:
Photos: from UAV with GPS data,...

Geometry

• Object type:Height field
• Geometry type: Smooth
• Quality: high - ultra high
• face count:0

My processing steps are:

• aligning photos
• building geometry
• export DSM
• building texture
• export Orthophoto
• export Report

I need a huge amount of "trias" for the DSM.
If i process the geometry and texture with batch then i see also the huge model in PhotoScan. Is this correct ? Or do you mean to write my own python script ?

Best Regards Marcus

[Wishgranter]

see this in the BATCH menu - attachment. so set it align, dense, mesh, ( DECIMATE to some  aceptable size ), EXPORT MODEL.. then can set decimate again to say 20mil for inspection in pscan. geenrate textures etc.....

[mwittnebel]

I think you use a different version of PhotoScan.
We using at this point in time 0.9.1-1714 (64bit).

With "mesh" you reducing the amount of faces ?
doc.activeChunk.buildModel(object="height field", geometry="smooth" ,faces=20000000)

I would like to have a huge amount of faces for a precised DEM/DSM (digital surface model).
For Example: 5cm/pix for an area of 1-5 sp km.

It would be nice if i could process this data without loading this visual result in PhotoScan.
(This should theoretical solve my problem ? )

[Wishgranter]

yes.. so can decimate the amount of triangles.... try calculate real need of tirngles per m2 then per km2

[mwittnebel]

face count = amount of triangles ?

Thx for your helping

It would be nice if PhotoScan could run in a command-line mode without this GUI.
(Or rendering the model depending on the zoom level, like Google maps .... this would save a huge amount of resources of the GPU for large models)

[Alexey Pasumansky]

Hello Marcus,

To avoid the model visualization in GUI you can do the following:
- before running Build Model in the desired quality reconstruct it using Point Cloud quality option (it will take a few seconds), then switch to the point cloud view mode, then run geometry reconstruction -  I think it will do the trick and view mode would not switch to the geometry after reconstruction is finished.

[mwittnebel]

is it possible to change the view mode via a python command ?

[Alexey Pasumansky]

Hello Marcus,

Unfortunately, not.

[mwittnebel]

maybe in of the next Versions or to XMAS ?

Now, its a bit hard to build an python script to run the processing automatically. (with my specific requirements)

Thanks for all the helping.

[Alexey Pasumansky]

Hello Marcus,

We're not sure it will be reasonable to include view mode selection through Python, but we plan to add "Disable view mode autoswitch" option to advanced preferences tab. I think it will implemented in the version 1.0.0 quite soon (but not in the nearest update).

[mwittnebel]

A "Disable view mode autoswitch" is a good idea.

I believe personally, that a view mode change via python, is a solution with less programming effort on your side. (comparison to a "Disable view mode autoswitch")
This would also increase the possibility to create more customized scripts to automatize the image processing.

Best Regards Marcus