Hi, I’m part of a team who uses Drone Deploy for flight planning in a cartographic boundary project, and it covers a considerable amount of area in South America. Capturing the images hasn’t been a problem at all so far, however we’re going through a hard time when it comes to processing them on Agisoft Photoscan Professional Edition (version 1.2.0 build 2152, 64 bit), because it’s running on a workstation which is too good for a home computer, and too unfit for such a tough task.
Our first attempt covered an area of 4,25 km2, photographing each square kilometer at a time because of the drone battery limit. So we would merge all these separate captures in a whole picture of the island in the end. When we imported the files on Agisoft to process it, the estimated time was around 150 days! And we could not do it in a single shot - each flight had to be processed separately, so we could join the images together at the end. Obviously that means we are prone to imprecision even having control points set in every flight, but it was the only way we could find to work with the available equipment.
The next step is doing the same work with a much bigger and detailed area (17,42 km2), which is set to last 400 DAYS to process once we take all the pictures we need. More than a year of work, and we pretty sure we can reduce that estimative with proper infrastructure. Maybe even achieving to do all the processing one time only. That means we are trying to find out specifically what kind of workstation we should buy aiming a better performance, considering our resources are around US$19,000.
Here is our current setup. It’s not a brand new but it’s all we’ve got. Any advice on what to buy is welcome.
- Intel Xeon CPU E3-1225 v5 @ 3.30GHz 3.31GHz
- RAM: 16 GB
- Windows 10 Pro 64 bit
- NVIDIA GeForce GTX 1050 Ti
Thanks in advance!
My setup works well with Metashape (new name for Photoscan):
Intel i9 9980Xe 18-core processor @ 4 GHz water cooled
128 GB of 3200 MHz DDR4 memory
Nvidia GTX 1080 ti
4 TB of M.2 PCie 3.0 x 4 SSD with 3.2 GB/sec read 2 GB/sec write
This system costs much less than half of your 19k budget.
You need more memory and processor cores and CUDA cores in the GPU to run faster. And a fast clock on the CPU. This combination could speed up your processing 3X or more. To know for sure, if you send me a link to the photos for a 1 km2 sample, I will process them and tell you how long it takes. What do you want for output? Do you want to include GCP’s in this test? How many photos for a 1 km2 site?
Wow, common specs… Is that from SpaceX?
@raphaelfp It really depends on what resolution you are willing to deal with. You could use a batch editing photo software like ACDSee to slightly knock down the images which will both upload and process faster, but there will be a slight degradation in the final product. Though it won’t be quite as bad as using turbo upload…
His problem is not the upload time. It is the small memory which forces Photoscan to work off disk much of the time. It is the processing time that is killing him.
I understand that, but it sounds like he is not using DroneDeploy for processing and probably on a free account for flight. The method I mentioned applies to processing regardless of software. I often process in Pix4D as well and have had trouble with really large projects running out of memory with 64GB. Optimizing the photos to 75% has worked in my case. It would probably be best to go to the AgiSoft forum for better coverage. Not that you aren’t capable…
Hey there! Thank you for your tips, they are really useful! I’m in touch with the drone team so I may find a way to share a sample, however GCP is not an option for the moment so I’m working on it. Each km2 contains 1329 photos. We are aiming for minimum precision of 5 cm and our control points are obtained with dual frequency GPS (L1 and L2).
You’re right, we’ve been using only a free account for flight and then stuck to the software we’ve got already. This was the first forum I was told about the subject so that was the reason I came here. We just can’t knock down the images too much due to the nature of the project, but you’ve got a point and I appreciate your advice as well I’ll be discussing all the info with the rest of the team and let’s see if something new comes up.
Send me a link to a set of photos and I can get started.