Software assessment

[ScottH]

I am having a hardware issue in terms of processing time and RAM usage. I have included an attachment outlining process steps, CPU and RAM usage and processing time. Any advice would be appreciated.

Thanks
Scott H

[Exhale]

Hi,
Quadro 600 GPU  and  807 photos ?  It sounds crazy to me..

I recommend you to have a look at this link first. http://www.agisoft.ru/wiki/PhotoScan/Tips_and_Tricks#Recommended_Hardware
This is for more info.
http://www.agisoft.ru/forum/index.php?topic=1683.0

Thx.

[Alexey Pasumansky]

Hello Scott,

1) Faster CPU will speed up all the processing stages and one of the recommended GPUs will also significantly increase the processing speed during Build Dense Cloud stage.

2) PhotoScan doesn't use all available RAM, but only the amount that is required to complete the current operation. The link to the Wiki page where approximate peak data depending on the number/resolution of images and processing parameters is given above.

3) 8 hours for 800 images is too long processing time. I've just ran a small test for Align Photos stage (about 830 images, High accuracy, Ground Control pre-selection, 40000 point limit) and it took significantly less time on i7 4960X, 3.60 GHz - the whole stage (Align Photos) has been completed in 1 hour 40 minutes (Matching time about 1 hour 10 minutes and 30 minutes for alignment time). Are you sure that Ground Control pre-selection wa used?
If you have the log file saved for this project please send it to us so that we can check which steps took the longest time. Also you can post Chunk Info screenshot for additional information.

4) Optimization is highly recommended if Ground Control Points are present for the project, since it helps to improve the georeferencing accuracy and fix possible model distortions. Yaw, pitch, roll data are not used during processing (unless you have very accurate camera GPS device and are using Camera Correction section in the Ground Control pane settings dialog).

[ScottH]

Hi Alexey,
Thanks for your reply. I have included attachments: 1) Screen capture of chunk information, 2) Log file (Zip).

In response to your reply
1)   Changing the CPU is not an option at this point. I am however prepared to install an alternative GPU. Although one of the recommended GPU is a NVIDIA type, it does contradict many of the posts I have read that states NVIDIA does not support OPEN CL as well as other models. I believe this should be addressed in the recommendations.
2)   Based on the charts in the recommendation section is why I purchased 32GB only to find it will never be used. There is some discrepancies between recommendations/charts and posts.
3)   Yes, Ground Control Pre-selection was used. (WGS84 – GPS equipped UAV).
4)   Not sure I understand this response. For this project,  ground control is an inaccurate onboard GPS – WGS84  (UAV collected). It does collect x,y,z, Yaw, Pitch and Roll. I also use a camera calibration report created from Agisoft lens.
a.   Are there any advantages to using the Yaw, pitch roll in the CSV data.
b.   I do a camera calibration report for each flight. Are there any advantages/disadvantages in using this report on flights conducted with and without ground control points.
c.   When camera calibration is used and I run the optimization I do not have the ability to ‘check’ any of the correction factors boxes, they are all greyed out. Is camera calibration a necessary step.

Any help you can offer will be appreciated.   Thank you

[Alexey Pasumansky]

1) I'm not sure which posts you are referring to. Recommended cards like GeForce GTX 580, GTX 680, GTX 780 and GTX TITAN are good for PhotoScan Processing. While Quadro cards are really significantly slower.

2) Approximate memory consumption peaks mentioned on our Wiki pages do not contradict memory usage you have based on the reconstruction parameters used and actually are in accordance of it. 32 GB will allow you to process several thousands of images in Medium quality.

3) Well, since CPU can not be changed, using low frequency quad-core processor will result in long processing time. The only possibility to speed up the process in this case is to use lower value for Point Limit (in Align Photos dialog). By I would not recommend to go under 20 000 limit.

4)
a) Yaw, Pitch and Roll angles are not used in the processing, the only case they are used is using Camera Correction section. But since it will not be very util for low precision GPS device, then camera orientation angles are not actually required to be imported in Ground Control pane.

b) If the calibration is accurate and no camera parameters has changes, then camera calibration can be used in Fix Calibration mode. Also it will be helpful if no GCPs are available.
But in case marker-based optimization will be used and focal length with sensor pixel size (both in mm) are known it is not really necessary to use calibration data, since PhotoScan performs auto-calibration procedure quite well and intrinsic camera parameters may be adjusted during optimization.

c) If you are using pre-calibrated camera with Fix Calibration option enabled in Camera Calibration dialog, all intrinsic camera parameters are locked and will not be refined, however, based on the camera/marker accuracy settings (in case any GCPs are used for the project) camera positions as well as points in sparse clouds positions may be refined during optimization.