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 - akemono

Pages: [1] 2
General / Re: Aligning huge projects
« on: August 03, 2015, 11:43:51 AM »
Hi Igor,

Yes i meant using the Pair pre selection Generic. Sorry i should have clarified this. In my experience the alligenment is not worse or better if using the pair pre selection. This does however applies to a aerial data set with plenty of overlap.

I did notice however with some data sets that have less ideal overlap or difficult geometry that disabling the pre pair selection helped in getting more images aligned. But this is more a sign of that either need to add more pictures, get more overlap or re-shoot it entirely.

Conserning the speed increase, it can be pretty drastic, 10x or more is possible (i think). Also depends on your dataset. It takes relatively long to compute the pre selection but after that the actual alignment is really quick. Have a look below at the results i got using a aerial dataset.

Total cameras: 4.000
Effective overlap: 6.7
Reprojection error: 1.4

Alignment paramters:
Accuracy: Medium
Pair preselection: Generic
Key points: 40.000
Tie point: 1.000
Matching time: 14 hours 40 minutes
Alignment time: 9 minutes 37 seconds

I'll do a quick test with a smaller datset to give your a idea of the time savings that can be had using the prepair selection. I will update this post with the results of that test.


Dataset with 38 35Mpx images

Pre pair selection disabled:
Camera's matched 38/38
Effective overlap: 5.46
Reprojection error: 0.96

Alignment paramters:
Accuracy: Medium
Pair preselection: disabled
Key points: 40.000
Tie point: 1.000
Matching time: 3 minutes 4 seconds
Alignment time: 1 seconds

Pre pair selection enabled:
Camera's matched 38/38
Effective overlap: 4.5
Reprojection error: 0.84

Alignment paramters:
Accuracy: Medium
Pair preselection: Generic
Key points: 40.000
Tie point: 1.000
Matching time: 1minutes 30 seconds
Alignment time: 1 seconds

New dataset with 301 Camera's

Pre pair selection disabled:
Camera's matched 301/301
Effective overlap: 8.98
Reprojection error: 1.28

Alignment paramters:
Accuracy: Medium
Pair preselection: disabled
Key points: 40.000
Tie point: 1.000
Matching time: 2 hours 26 minutes
Alignment time: 27 seconds

Pre pair selection enabled:
Camera's matched 240/301
Effective overlap: 5.94
Reprojection error: 1.1

Alignment paramters:
Accuracy: Medium
Pair preselection: Generic
Key points: 40.000
Tie point: 1.000
Matching time: 19 minutes 25 seconds
Alignment time: 31 seconds

Oke so with Pre pair selection enabled it's a lot quicker! Using the same data and the same parameters, it's 7.3 time quicker then disabled (2 hours 26 minutes = 146 minutes -- 146/20=7.3)
There is only one drawback i found, using the pre pair selection enabled it would match 240 camera out of the 301. And disabeling the pre pair selection it would match all of the cameras. This is probably due to the outer camera's not having enough overlap with the rest of the camera's.

So there you have it, it's clear that it is a lot faster but it might not solve all of your camera's. (because of insufficient camera overlap at the edge of the model.)

General / Re: Aligning huge projects
« on: July 29, 2015, 04:39:43 PM »
We actually had a similar problem. We had a project for aerial visualization. We ended up taking 15.000 pictures. And we also tried to align them all in one big chunck, wich failed because we ran into a shortage of RAM.(i use a pc with 64GB of RAM)

Luckely we could use every second image cutting the total aligned images in half. This did fit in the RAM.

To dramatically speed up the alignment process turn on "pair preselection". This feature enables Photoscan to generate a low resolution version of your image and pre compare these for matches. If those are calculated Photoscan continues on to compare only those images that are in the vicinity.

Hopes this helps a bit.
Best mike

The quickest way to fix your messed up photo locations is reload a previous save.

If you however live in the real world where you don't save every second, you might have to realign the entire dataset.
However! In the Reference tab you can also uncheck all your GCP's and check all your camera's and "Optimize Camera's". This might return your camera's to their original location.

At work I also have the delight to work with 2 coordinate systems. After some searching i found that you can tell Photoscan to ignore the coordinates the camera's have and just use the GCP's. To achieve this go to the Reference pane, and disable all the camera's. Then enable all your GCP's and Optimize your camera's. This works for me.

Hope it helps you a bit.

General / Re: Texture Export question
« on: March 31, 2015, 12:46:03 PM »
Just adjust your max materials. If your using the standard material set it to 100% Self-Illuminating. This should fix the problem your having.

General / Re: Import into Blender
« on: July 10, 2012, 01:07:00 PM »
Hi guys,
Importing it into blender is not that hard. Getting the texture to show can be a bit of a hassle.

If you just want blender to show you your texture in the view port try the following:
  • Import your mesh
  • Select your mesh and select edit mode (or press Tab to enter edit mode)
  • Then at the top of your screen you can see, next to help, Default. Click there and select UV editing
  • In the uv editing mode go to the tab that says 'Image' and search for your texture file
  • After this go back to default and go back to Object mode
  • And finaly press 'ALT Z'
Now your should see your textured mesh. If this isnt working for you let me know, i will try to explain it better.

Best Mike

General / Re: PDF Limitation on Texture
« on: April 23, 2012, 10:24:35 AM »
That looks really impressive, it has a high quality feel to it. I think they had to optimized the meshes/textures to achieve this kind of performance.

The people who created this obviously know their way around the Cryengine.

General / Re: PDF Limitation on Texture
« on: April 17, 2012, 04:06:07 PM »
Hi Oli63,
Unfortunatly i did not work with the Cryengine yet. But that might change in the near future.

Developing a game is not that much different then from making a model. You have your mesh that you import into the game. Though it usually has to meet certain criteria. After a succesfull import you can apply a shader to it and assign a texture to that shader. Then add some way of moving the camera and perhaps a HUD and your done.

It's really not that hard, youtube has many hours of tutorials on this stuff.

But there are some "hard" limitations, for instance:
  • Unreal can handle up to 8k textures
  • Unity can only handle up to 4k Textures
  • I don't know Unreals poly limit per mesh
  • Unity can handle up to 64K vertexes per mesh, after that it offers you the option to divide the mesh up into smaller parts.

So most of the time your limited by how big your textures can be.
Insidious Technologies has a very nice plugin for Unity that allows you to use up to 512K textures. It's super cool feature but the only drawback would be that you can't use it online.

To view your results in a game-engine you have to bake the game. This basically tells the game to package the game and compress it into something more convenient to transfer. After this process you have a installer that will install the game and you can run it like any other game. This way you can show it to potential customers.

Alternatively you can also run it "live", UDK Cryengine and Unity all support this. In the editor you have your scene and you can play it from there. However this has one drawback, you have to run both the Editor and the game at the same time which is more resource intensive and does not always give you a smooth playback in all circumstances.

Pffew, i think thats enough info right there. If you want to know more just pm/mail me and i'll try to help you the best i can.


General / Re: PDF Limitation on Texture
« on: April 17, 2012, 02:44:02 PM »
I tried it but i think that fibix is only usefull if you want to render offline. I didn't find any online examples, plus the installer needs to many clicks to install for the average user to be interesting.

I still think that Unity or Epic will be a better choice, that is ofcouse if your main goal is to show it on the web. Epic has some form of Flash support built into the Source code licensed version of their game engine. Unfortunatly i can't check it because i'm not going to pay X$ to just find that out. Their UDK does not have the Flash support, but maby that Flash will eventualy find it's way to the UDK as well.

Try a game-engine like the ones I described above, and you can also consider using the Cryengine. Im sure that will blow your customers away ;)

General / Re: 3D models on websites
« on: April 11, 2012, 10:09:33 AM »
Well here is a result using P3D. The mesh part looks good, the only downside is that the texture space is extremely limited (only 1024*1024). I haven't looked into three.js yet but that will be my next attempt.

Here is a example of P3D

General / Re: 3D models on websites
« on: April 03, 2012, 11:00:57 AM »
Mark that looks really good!
I totally agree with both you and Wishgranter that we need a different approach to display our content on the web then just using the PDF file.

Personally i have high hopes for Unity because of it relative low cost, its ability to run on low end hardware, and it's many export options.

I have worked in the UDK for more than a year but that was mostly for game design stuff at school, and i haven't touched it in the last year or so. After that i worked for half a year with Unity this also for a game that i was developing with small group of students.

Perhaps we can work together to reach a consensus on what would be the best way of getting our models on the web.


General / Re: Graphics cards and memory
« on: April 02, 2012, 05:54:44 PM »
Hmm oke, thanks for the quick reply.

To bad that it is hardly any quicker then the 580 GTX, but thanks for testing it!

General / Re: Graphics cards and memory
« on: April 02, 2012, 03:45:45 PM »
Alexey did you already have a chance to test the new ATI 7970? Im quite interested to see what kind of speedup that would provide vs the current Geforce 580 gtx.


Face and Body Scanning / Re: Scanning people: Tips needed
« on: April 02, 2012, 03:41:47 PM »
Hi Mark,

I think your model looks good, and i don't just mean the woman in the pictures ;) Only thing that stuck me was the lack of high resolution textures. That can help tremendously in giving a more realistic feel to your work.

As for your problem with the subject moving, you can try and setup a kind of dome with a few dslr's equipped with wide-angle lenses to cover as much area without to many camera's. Still this setup would cost quite some money but that is probably the best way to go if you want to capture your model in one go. As for your problem with simultaneously triggering all those camera's, depending on the type of camera, you can get a Infra red remote that will (probably) trigger all the camera's at once.

If you want  to discuss it further or need some other help, you can send me a mail.


General / Re: Recommended Hardware?
« on: March 30, 2012, 04:08:20 PM »
Sorry to bump this but i wanted to follow up on how our new hardware performs.

The new systeem preforms at least 3 times faster then my old systeem! Which is incredible!
And there is even room for improvement, we could get another GTX 570 which in turn would make the systeem 50% faster still!

They only downside is that even with 64Gb of ram i would still get a "out of memory" error when i try to process 84 16mpix images. Usualy this happens at the "creating mesh" stage after the depth calculation. Is this normal?? Because i thought that 64Gb would be plenty to use the software to it's full potential....


General / Re: 3D models on websites
« on: March 12, 2012, 02:04:30 PM »
@ Wishgrater,
I saw your demo using unity, nice preview but i see that your also having issues using high res textures with unity.

Im having the same problems, i need to import a large model, half a city, into unity. No problems on the mesh side of unity, it runs perfectly even with 2 milj polys, but i can only use one 4k texture.

A possible solution could be to cut the mesh and calculate a new texture using Photoscan but this would require very preciece cutting that is not doable in PS. Any ideas?


Pages: [1] 2