Topics

16

General / Running Photoscan on a VM - another NVIDIA driver rollback

« on: August 25, 2014, 08:49:50 AM »

As mentioned in the benchmarking thread I'm testing PS on a VM provided by our IT guys. The first test with 8GPUs and 8/16 CPUs threw up some GPU errors e.g.

Code: [Select]

Reconstructing depth...
[CPU] estimating 1240x756x160 disparity using 1240x756x8u tiles, offset -138
[GPU] estimating 1399x873x288 disparity using 1399x873x8u tiles, offset -209
ocl_engine.cpp line 320: clWaitForEvents failed, error -9999
GPU processing failed, switching to CPU mode

(Full log attached.)  Can anyone suggest what might be causing this?  The end result is successful, there's a performance result for the device and it's faster than with GPU disabled and all CPU cores active. Just a timing alert?

Also, which bits of the log file are used for the benchmarking?  I know it's not going to be the fastest thing around but it would be useful for us to know where this fits in with other hardware options for comparison.

[edit]Asking our IT guys for more details on the system specs as I'm guessing there may be some driver settings to tweak(?)  At the moment our Dell T5600 (K4000, 3Gb) is slightly ahead overall due to better GPU performance.

17

General / Does this sound practical? High res scanning of over-sized prints.

« on: June 03, 2014, 06:38:06 AM »

Hi All

We're currently digitising a collection of Piranesi prints in bound volumes (e.g. http://cat.lib.unimelb.edu.au/record=b2984624) and I've been looking at options for scanning large foldouts.  They're too big for our largest scanner (A1). I've considered using panoramic stitching software, but setting up the equipment to do that has its own challenges as the books have to be kept lying down, so the camera would have to be mounted from the ceiling.

As an experiment I thought I'd try creating a model of the foldouts in Photoscan and render that out at high resolution... but then I kicked myself. an orthophoto would be easier. I just need to get the pro version.  I'd appreciate some thoughts on the practicality of this and possible workflows.

• Would there be any advantage to using a wider angle lens to help align higher resolution images?
• Should I shoot much higher resolution than I need for the final output? Any suggestions for how much?
• How low can i go with the dense point cloud and mesh given that I''m mainly interested in just the texture

I'll be testing these things out anyway but I'd appreciate any suggestions/ideas.