I have not done it my self .
I have never tried running without Quadro card selected. Ā Will try a 2,000 pic dataset tonight. Ā
The 2,000 pic dataset has been running 24 hours and is 56% complete with step two. Ā Step one was completed, GCPās picked and reoptimized prior to starting this processing run. Ā The Quadro card is off processing settings are to use 126GB RAM and all 16 processor threads. Ā It is currently using 49GB RAM and all 16 threads are running at 100%. Ā Only the merged LAS pt file is selected for an output. Ā Does this seem like a normal time frameĀ for 56% progress from anyone elseās experience? Ā
For what itās worth Iāve run Pix4D with on several computers with Quadro cards. K600, K3000(?), single K5000, dual K5000, and a current system running both a GTX and a K5000 at the same time.
I did notice above that the output for the 2000 photo set is a merged las? File sizes are going to start becoming large depending on photo and output resolution.
If it hasnāt been mentioned, a question any user could ask would be whether a 1:1 point cloud gains them much over 2:1 or even less. Iāve made my share of very dense point clouds for various reasons, but itās a very large amount of processing overhead to carry through the production process that might not be necessary. It might not even be better.
Steve
Ryan - yes that sounds normal to me. Remember to clos eteh project afte rstep 2, re open and start step 3 so the Load Point Cloud doesnt hang. I spoke to P4D yesterday and they are working on the Xeon issues.
I have just started a 11,365 image dataset. Will let you all know how that goes and how long each part takes. Running on my dual Xeon E5 2620 v3 with 128GB RAM and the Quadro K5200 card. All latest drivers installedā¦Well I processed all of teh 20,000 images in 10 blocks - so I thought I would do it in 2 blocks and see what happens!
By the way, does anyone know of a good way to harmonize/standardize the contrast for an orthomosaic that was taken over different days and with different light conditions?
Ā
Phillip - Did P4D have a general time frame for improvements?Ā Ā Our Trimble software runs at least 3X faster (the bigger the project, the bigger the speed difference between Trimble and P4D) on same size and density cloud.Ā
Hi Ryan - no commitment on timeframeā¦
One thing to mention (in case you didnāt already know) about different density point clouds is that you shouldnāt loose any accuracy for the project. We did extensive testing on a project with 20 GCPs, and around 20 confidence points and found that low density got us the same results it just didnāt look as pretty visually.
Not that this was asked but it was mentioned by Steven that you may want to be careful of file sizes. Itās one thing to get Pix to process them, itās a whole different beast to actually classify and make them usable in any other software, Iāve foundā¦
Good luck, this sounds like an awesome project!
Thanks Jerry. Will keep a note of this. How does Photoscan compare in terms of processing speed? Have you tried this?
Direct comparisons between software can be a tricky thing. They donāt all work the same. Up until a couple years ago anyway, not all algorithms were even using individual colors but just overall gray shades. Some were matching successive sets of stereo-pairs.
Some legacy program suites could handle massive data sets and did exactly what they were supposed to do, but didnāt produce the incredible detail possible in point clouds from some of the newcomers. The cloud was just a means to a DTM. Some of those programs werenāt able to work with images with any significant tip, let alone completely random photo locations and directions.
The issue with the Xeon not providing the boost it should is unfortunate. But there are many factors to take into account when comparing software.
Iām curious as to what end product most users are working toward. Hi-resolution orthophotos? Volumes? A cleaned terrain model? Are a lot of folks actually using point clouds down to the pixel level? Make no mistake, on the right projects I want to be able to make them as efficiently as I can(!), but unless that is the desired end product its a staggering amount of calculating to perform and a lot of overhead.
Philip, I donāt mean to discount your question, just make a point that Iāve seen much variation in the function and products between various programs.
Steve
Heres an update on the large dataset:
The 11,000 image dataset crashed/locked up during Stage 1, Calibrate.
I therefore reduced the dataset to 8500 images.
Time to complete 8500 image dataset Stage 1 only using dual Xeon E5 2620 v3 (total 12 cores running at 2.40 GHz)/128GB DDR4 RAM/Nvidia Quadro K5200 with 8GB RAM: 24 hours, 21 mins.
For those that think the Quadro card is the issue, I have full logging enabled with all headers and I noticed that whenever a task is passed to the GPU (for example the matching pairs in stage 1 when the blue dots move like lightning) there is no slow down at all. In fact as mentioned, its like lightening. So I am not convinced the Quadro is the bottleneck. I donāt have another card to try unfortunately - but I almost pressed the Buy button for the GeForce Titan Xā¦
I spoke to P4D the other day, they are fully aware of Xeon issues but are not able to identify the issues with some of the newer Xeon chips and why they are causing a slow down.
My view is that Xeon is becoming more mainstream. Multi eBee projects allow us to gather huge datasets never before possible. For example my project is 20,000 images that I collected over 4 mornings here in Dubai (too hot after 10:30 am). Datasets are getting larger, users are demanding more from the Mapper Pro app. I am confident that P4D will resolve this - just not sure when.
Ā
Ā
Ā
Philip is there a way you would be will to send me the data set to compare it in my system?
Not sure how i can get such a huge amount to youā¦WeTransfer maybe? What system are you running?
Dual Xeon . I give you access to my FTP and you can upload it through FTPzilla client.
I think you will need to send me the login details
I have it. Whatās your email?
What version of Windows 10 Pro is everyone running?Ā I noticed that after updating to Version 1607 OS build 14393.10 I started to have issues (it is hanging) again with step 2 and step 3.Ā I have not completely narrowed it down to this because I also installed a second GTX970 card.Ā But it will not be hard to figure out if the update or extra card is the problem.
Latest version of Win 10 for me, whatever that is.
You can check it by typing āAbout your PCā in start menu
Ā