DIY and Entry-Level 3D-Scanner Forum
http://www.diy3dscan.com/

Test with photoscan
http://www.diy3dscan.com/viewtopic.php?f=42&t=118		 Page 1 of 4
Author:  sambo3D [ Thu May 23, 2013 3:28 pm ]
Post subject:  Test with photoscan
So, I would not call this a project I am proud of but still it is a test.

I used ambient light on the first attempt and a ring flash for the second one.
In both cases I have moved around the object for 26 times. Used a 50mm Carl Zeiss lens on a Canon 5D Mark II Body. 100 ISO, f4, 1/15. In both cases.

She is Saint Lucy.

Low level geometry, 5000 faces (I know, very low specs + eager to see some results in less than an hour).

My machine? A Macbook pro, dual core, 4gb ram, gt 9600 512 RAM + gt 9400 256 RAM hence 6 GPU cores.

6/6 GPU CORES USED
0/2 CPU CORES USED

hope might interest you.

One thing interests me in particular: the "debris". Using a ring flash somehow accentuates the difference in exposition between foreground and background, a difference that becomes a good excuse for the software to create nice meteors flying around...

I am using the demo version, hope it is possible to post screenshots here.

Bye!
A.

Attachments:
1_3.jpg
1_3.jpg [ 25.7 KiB | Viewed 39884 times ]
1_2.jpg
1_2.jpg [ 46.41 KiB | Viewed 39884 times ]
1_1.jpg
1_1.jpg [ 44.98 KiB | Viewed 39884 times ]
Author:  sambo3D [ Thu May 23, 2013 3:29 pm ]
Post subject:  Re: Test with photoscan
Other two images from the first attempt (ambient light only)

Attachments:
1_5.jpg
1_5.jpg [ 39.63 KiB | Viewed 39884 times ]
1_4.jpg
1_4.jpg [ 25.23 KiB | Viewed 39884 times ]
Author:  sambo3D [ Thu May 23, 2013 3:30 pm ]
Post subject:  Re: Test with photoscan
images from the ring flash attempt. ring meteor result

Attachments:
2_3.jpg
2_3.jpg [ 42.31 KiB | Viewed 39884 times ]
2_2.jpg
2_2.jpg [ 47.45 KiB | Viewed 39884 times ]
2_1.jpg
2_1.jpg [ 26.02 KiB | Viewed 39884 times ]
Author:  sambo3D [ Thu May 23, 2013 3:31 pm ]
Post subject:  Re: Test with photoscan
very last one.

Attachments:
2_4.jpg
2_4.jpg [ 50.19 KiB | Viewed 39885 times ]
Author:  virtumake [ Thu May 23, 2013 5:01 pm ]
Post subject:  Re: Test with photoscan
I am impressed by the results of your 5k polygon model. The textures are great.

Did you move around the object, or did you put it on a turntable?

I think a big advantage of this system is the scaleability. If you have a macro lens, you could scan tiny objects as well, I guess. Unfortunately I do not own a Canon to test it by myself :(

Do you think you could have achieved these results with a cheaper camera as well?
Author:  kiesel [ Fri May 24, 2013 9:46 pm ]
Post subject:  Re: Test with photoscan
Hi,

I have used the Standard version for 2 and a half year now and I'm still impressed. You don't need a DSLR to work with it. For makro work point-and-shoot cameras with their bigger field of depth are even better. One advantage of photogrammetry is you can use it were ever you are, for example on your holidays ;) . On my Cornwall trip in 2011 I have shooten an old cross of stone. Some results are posted on http://www.pgrammetry.com/forum/viewtopic.php?f=48&t=547&start=10 .

Cheers,

Karsten
Author:  virtumake [ Sat May 25, 2013 7:45 am ]
Post subject:  Re: Test with photoscan
Nice scan :) I downloaded the demo and my laptop is working on a scan of a figurine. It says it needs an hour to do so.

I recognized that rhe 3d model is cut in the middle. Is this because of the demo version of photoscan?
Author:  sambo3D [ Sat May 25, 2013 8:26 am ]
Post subject:  Re: Test with photoscan
kiesel wrote:
Hi,

I have used the Standard version for 2 and a half year now and I'm still impressed. You don't need a DSLR to work with it. For makro work point-and-shoot cameras with their bigger field of depth are even better. One advantage of photogrammetry is you can use it were ever you are, for example on your holidays ;) . On my Cornwall trip in 2011 I have shooten an old cross of stone. Some results are posted on http://www.pgrammetry.com/forum/viewtopic.php?f=48&t=547&start=10 .

Cheers,

Karsten


Hi Karsten, it is amazing, I used photoscan for one week and to me it is still magic.

I wanted to ask you (as also Bernhard was asking this): is there a sort of relationship between what you shoot and the lens you use?

First example: using a macro on a small statue, producing, say, 1000 images, will photoscan be able to align them? It has been said that a masked object is the best scenario for photoscan. When I use a macro, though, the whole image is occupied by a small portion of an object, there are no reference points (at least to human eye). Will PS be able to reconstruct? Maybe he needs at least 1 to (say) 10 pictures of the whole statue to have a reference?

Second example. I need to "scan" a table with people eating around it. Do I have multiple ways to achieve my result? Shall I use a macro for 200.000 times, mapping every single square centimeter or shall I use a wide/50 lens? I tried yesterday shooting a table, I shot 63 images with a 50mm and I was no more than 50 cm far from my object. I used a tripod and moved around the table.

Eventually, you say it is the same using a non DSLR vs DSLR, would you agree, though, that a Carl Zeiss lens is way better than a not weel-defined zoom lens on a 100$ camera? In terms of image definition: clarity of the actual pixel, which would be good for the point cloud, and the sharpness of the textures.

thanks!
a.
Author:  sambo3D [ Sat May 25, 2013 8:41 am ]
Post subject:  Re: Test with photoscan
virtumake wrote:
I am impressed by the results of your 5k polygon model. The textures are great.

Did you move around the object, or did you put it on a turntable?

Do you think you could have achieved these results with a cheaper camera as well?


I am rotating around the object with a canon mounted on a tripod.


I do not know yet... but yeas, as Karsten says, you can achieve the same results. Still I am investigating on the relationship between different lens quality.
Author:  kiesel [ Sat May 25, 2013 4:01 pm ]
Post subject:  Re: Test with photoscan
@Bernhard,

this is my second answer after I lost my first because of automatic logout. :?

Bernhard:
Quote:
I recognized that the 3d model is cut in the middle. Is this because of the demo version of photoscan?


No, there is no difference in functionality.

Sambo3D:
Quote:
Is there a sort of relationship between what you shoot and the lens you use?


Yes of course, you should choose your lens (and camera) in dependencies to what you want to shoot. It is a question of image scale.

Sambo3D:
Quote:
First example: using a macro on a small statue, producing, say, 1000 images, will photoscan be able to align them? It has been said that a masked object is the best scenario for photoscan. When I use a macro, though, the whole image is occupied by a small portion of an object, there are no reference points (at least to human eye). Will PS be able to reconstruct? Maybe he needs at least 1 to (say) 10 pictures of the whole statue to have a reference?


I think 1000 images are way too much for this. I think you need 24 images for every 360° turn around your object. Begin with pictures of the whole statue and then add some detail shots. Keep always in mind that you have to shoot stereo pairs because it is a multi stereo approach! Every part of your object should at least shoot with three images. Nevertheless Photoscan should be able to do this in according to http://www.agisoft.ru/wiki/PhotoScan/Tips_and_Tricks it needs 5 GB memory for alignment of 1000 images. You need masking if your object is moving against its background (turntable solution). Even then it is sometimes helpful if you choose a turntable table with a good texture (a newspaper, cork) and doesn't mask it out.

Sambo3D:
Quote:
Second example. I need to "scan" a table with people eating around it. Do I have multiple ways to achieve my result? Shall I use a macro for 200.000 times, mapping every single square centimeter or shall I use a wide/50 lens? I tried yesterday shooting a table, I shot 63 images with a 50mm and I was no more than 50 cm far from my object. I used a tripod and moved around the table.


200.000 phots are you kidding me? :D What do you mean with "... a table with people eating around it." Do you want to scan the people too? For this you need a multi camera rig bigger than this by Lee Perry. The table alone should be not the problem, if it is made of wood with good texture.

If someone have a good photo set, with he/she is satisfied I could make the model of it to overcome the limitations of the demo version.
Otherwise it is possible to get a model with Autodesk 123D catch http://www.123dapp.com/catch . Much faster because it is a cloud solution but you lost your copyrights of the images and the model.

Cheers,

Karsten
Page 1 of 4 All times are UTC
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/