Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - Artisan S

Pages: [1] 2 3
Hello Artisan,

Can you provide the related log and  specify the configuration of your system: OS version, CPU, GPU and driver version?
Windows 10 Pro latest build...
MSI X99A mainboard (1.09 BIOS is memory serves)
Videocard MSI HTX980Ti driver version.....382.05
32 Gb of 2166 memory

Attached an image of the final screen......inculding the log.....the real log is lost......but this is (and that is good) happend to mee three a row. You see  the videocard to a hollyday and the processor was running nearly idle....only IO was rampant (I waited for an hour).

Workaround for me, I reverted to 2.6.1......waiting for a solution.

Greets, Artisan.


I got it working. I reverted to 1.2.6, stuck my computer in a Tardis to speed up teh procedures and started calculating. Great.....but since teh Tardis was reconfiscated by a guy called The Docter (which to me was Valentino Rossi) I had to revert again to 1.3.2. So I reinstalled the whole software. And I did not use the MSI software Afterburner......and everyting worked like a charm without a I took 250 of the best pictures checking overlap and let it rip for a high DPC. No problems and a 6.5 million faces mesh, which in resolution outshines a scan made with a 60K Creaform scanner. Now I must say the Creaform scan was a lot faster but qualitywise it's no match anymore.

Keep up the good work........


Same problem 1.3.2 (downloaded and updated today)  with a job of 445 photo's in mode medium and even in lowest mode for dense point cloud generation. It seams to grind to a halt when it needs to preload the data at 89% of the point cloud generation job. Never seen this before. What is strange is that:

a) The times has run down to 0 shortly after the proces hit 89%........

b) The diskload is around 50% continous and that is a pure read writes.....

Memory is 32 Gb and the load never rises over 40% even in medium load.

Greets, Artisan

General / Re: Different scale for different chunks, is that normal?
« on: January 06, 2016, 02:58:02 PM »
Yeps, I get that (now).....sonud like logical....verry logical indeed. I had the same view in both chuncks (so the contents were overlapping a bit) but not the photo's, thanx Alexander.

Greets, Ed.

General / Re: Different scale for different chunks, is that normal?
« on: January 03, 2016, 10:50:59 AM »
Okay, that I was affraid the situation is simple, provide enough overlap and hope chunks join (so use overlapping camera ranges for instance). If this does not align using points, try using overlapping camera's to join (it works more of less) and if that does not work, buy the pro-version and align the chuncks by hand. Got that  ;D.

That somehow will cause some problems for me, but I'm confident I can find a pipeline using other software that will fit my bill. So the first thing I have to do is to solve the scaling problems by externally scalling AND aligning the pointclouds. That does not look undoable  ;).

Greets, Ed.

General / Different scale for different chunks, is that normal?
« on: January 02, 2016, 09:12:00 PM »
Problems aligning chunks....

I made 2 chunks a set of photo's of a Raspberry fore the topsise circle and a sideview circle and on from the other side. Now I alligned the chuncks using the point method (10.000 points, use masks the hole nine yards, so to speak). Result no alignment but what was really suprising also no match in that seems strange to me, I use the latest version of the standard software. Is this a bug or am I stressing the software to much?

I got rid of the chunk and noticed that 78 photo's had also done the trick.....even with a shiny object like a PI....

Greets, Ed.

General / Re: Sparse cloud holes problem
« on: January 02, 2016, 08:57:40 PM »
Indeed, try to get each surface under and angle of around 35 degree's for a real good oublique scans lead to holes in the mesh....

Greets, Ed

I don't think the OS does make all the difference, a freind of mine owns a 64 Gb 2 x Xeon Apple with some K cards of mr. Nvidia.....same results, only a bit later.

Greets, Ed.

It does not's polite Russian will inform you that you can save your project if you want to, but it will stop the calculations. Yeps I looked quite puzzled when that happenend.....but usually I can see the memoryload creeping towards my memory limit (32 Gb) and then I stop Agisoft and start using chuncks and allign then in other software that is a bit less memory hungry like Cloud Compare. Now dense Point Cloud generation isn't the most memory intens step (mesh creating needs a lot more). But looking on this forum I know that my machine (though fast enough) does not contain nearly enough memory to do really big tasks like arreal surface reconstruction on a big scale.

Greets, Ed.

General / Re: Aligning single flight line, i.e. Pipe or road
« on: December 31, 2015, 04:02:28 PM »
Agree, I have no such problems since I use (bought) a well known lens, camera combination, so it usually callibrated....also you I clean up the sparse point cloud in such cases....using this method:

I can asure you, it works for me.

Greets, Ed.

General / Re: using symmetry to render a solid object
« on: December 31, 2015, 03:59:50 PM »
Agisoft can't manage that, no way, in fact symmetry can throw Agisoft of kilter. Logically, since how is Agisoft to know which photo belongs were, if they seem to be duplicats of each other. Then using pair-to-pair matching can sometimes save the day but then you have to use a verry ordered shooting pattern.

Greets, Ed.

General / Re: Big project Dense Point Cloud classification
« on: December 31, 2015, 03:54:56 PM »
That would speed up things....what you are atempting is a bit on the megalomanic side of seriuously megalomaniac. A 980TI is not a wonder weapon, for doing a job like that you need a Tardis to send you computer back to the 70th :-). 500 12 Mpixel images and a high point cloud took 24 hours on my 32 GByte 6 core 980TI machine. More memory would also be nice. And 2000 million points by 4 is about 8000 million points....every step amplifies the stakes by 4, so a 16 Mpixel photo on ulta-high uses every pixel, on high every 4th pixel (4 million x 11.000 x 3 or 4 bytes = a ^%$load of RAM) and time taken is always a bit (a verry big bit) of a guess. I've seen Agisoft construct a 200 million point cloud from 89 pictures in about 5 hours calculating time (both on GPU and CPU simultaniously) and I have waited overnight on a 4 million pointcloud from setting high using 227 frames from the same camera. Beats me....

But 11.000 pictures is a cardload.....djees!

Greats, Ed.

I've either been amazed about the abilities of Agisoft to align photo's and been frustrated about it's unabilities....when I can I shoot concentric circles around the object, either using the turntable method or not....depending on the object's size. I've noticed that overlap is important but Agisoft has had some problems with:

1) Chunks.....individually alligned chunks did not have the same scale meaning reshoot please...this could be my error hard to tell, and personally I'm solution oriented in my work approach, meaning that I don't linger on checking the why but focus on prevention in the future....I standardised my shooting approach as much as possible (see above).

2) Infil shots.....when you have a complex model (lets say a Matchbox Model truck) I like to take infil shots to pay extra attention on certain sight lines I would be missing in the concentric circles, well this seems to baffle Agisoft sometimes. So I stopped bothering with these....Agisoft has it's own logic for filling in holes, and it means that even with smal details caught (and masked out) I can still be confronted with some unwanted action form Agisoft.

3) Mask. I always mask, you means always, like in always? Yes.....always. I know it sucks and it's feels like real work, but it improves allignment and improves scans.  Especially white sky, blue sky and trees in front of buildings, waving trees are the arch enemy of any Agisoft scanner. So mask them out and shoot around will improve alignment. If can get behind them, if not well you're gonna end up with a partially wavy scan or a scan with some ugly holes.

4) Depth of field....aperture 22 is minimum for shooting macro (and a APS-H or C sensor is preferred over a FF) since DOF is bigger.

5) Don't use flash....unless you know what your doing. I use a ringflash (idea of a german paleontologist and those people are never wrong about anything :-)). Keep the ligth as even a possible, use a tripod and if needed one ore two fill in flashes just to white out the background. Then Magic Wand selection in masking is a breeze....

Does this get 100% predictable results, no. there are still nasty and wonderfull supprises. Like thos bottle opener....tried it with laser, no scan, tried it using SLS no allignment between back and front. With Agisoft first time right.

Greets, Ed.


Bug Reports / Re: can't export point cloud
« on: December 29, 2015, 09:30:35 PM »
Okay, thanks for the workaround, looks workable if it works.

Greets, Ed.

General / Re: OpenCL has vanished
« on: December 29, 2015, 02:12:49 PM »
I had the same issue after a major update of Windows 10 pro on my 980TI, I solved it with an upgrade to 359.06 and an upgrade to the latest Agisoft version. When both were on par my Open CL worked like a charm.  My OS build is version 1511 and build is 10586.36. Verry sloppy work of the Redmond tribe I must add.

Greets, Ed.

General / Re: Aligning single flight line, i.e. Pipe or road
« on: December 29, 2015, 02:05:33 PM »
Try it, in my experience yes, but it depends on the subjects you are shooting of course. If there is enough information in the various shots, why not.....more lines give more information of course but sometimes you have to make compromises and make do with what you've got.

Greets, Ed.

Pages: [1] 2 3