I am struggling with the amount of time image processing is taking, and wondering if mesh decimation step is necessary for my purposes. I thought that I needed to limit vertices to 2 million because of my memory, but I see in the processing that the mesh is created with more vertices fairly quickly (a few hours), then mesh decimation gets my memory use up to around 100% and sometimes takes ~40 hours (I think it steps into the hard drive and virtual memory?). This step reads:
>>>decimating mesh (109752237 -> 2000000)
I am producing orthophotos and DEMs, so if I can make the mesh without decimating it, I think the quality might be better. But if I try this, will it make manipulating the surface in PhotoScan really hard, or will I still have memory problems?
My chunks are ~150-200 12MP photos with no GPS data covering an area about 1 km x 10 km. My workflow is:
1) Align Photos
2) create geometry
3) add GCPs
4) refine photo alignment
5) rebuild geometry
6) export orthos and DSM
Thank you for any advice.
I am running PhotoScan Pro on a Dual Quad Core Xeon 2.93 GHz machine with 12GB of memory and ATI Firepro V5800. Next week will upgrade to 48GB of memory. Waiting for a good price on a couple of GTX 580 cards...