New alignment parameter for background suppression

[Alexey Pasumansky]

Hello Nightingale,

If you have a complete set of accurate masks, then it is reasonable to use Apply Masks to Key points option (works like former "constrain features by mask" option), the new option should be used if you have only a few masks applied that cover the background.

[Smallpoly]

Just tried out the new masking approach. This is awesome and removes so much redundancy! Really clean scan too!

Model: https://sketchfab.com/models/2ad8fa7c276b43e1a3d3cfa130d96dc9
Data Set Examples: https://imgur.com/a/JOjNgH5

[Alain]

Hi

The Banana-Set works fine for me.

But my own test does not work well.
In addition to the Backgroundimage I added a few more masks.

One thing confuses me:
Photoaligning with "Medium" accuracy uses all photos but photoaligning with "High" accuracy only uses half of the photoset.

Any idea what I do wrong ?

Kind regards
Alain

[Smallpoly]

I'm not associated with Agisoft, but I can take a look if you feel like PMing me your photo set.

Just from the photos, it would probably help the results to take the shot from the other side, so you're getting the more brightly lit side, but it could be plenty of other things too.

[Mak11]

This has never worked for me. Banana set works but mines simply never work which is really frustrating to say the least..I even tested it with exactly the same setup with my own banana and it doesn't work...it's maddening.

[Alexey Pasumansky]

Hello Mak11,

Maybe you can share any problematic image set and send it to support@agisoft.com, so that we could check what is wrong with the approach?

[Mak11]

Alexey,
After some more thinking I may have found the issue on my side. Will have to confirm it by shooting some more test scenes and see if it finally works now. Will keep you up do date.

[Mak11]

Hello Mak11,

Maybe you can share any problematic image set and send it to support@agisoft.com, so that we could check what is wrong with the approach?

Alexey

I've just sent you a mail with a photoset.

Cheers

[Alexey Pasumansky]

Hello Mak11,

Thank you for sharing the dataset.

I've sent you back the project file with the alignment results I've got following the approach described in the blog-tutorial.

I've used the following steps:
- added all images to the same chunk, including the background image,
- applied mask to the background image only (DSC07632.JPG) - mask covered full image completely,
- run Align Photos operation (High accuracy, Generic preselection, key/tie point limits 40 000 / 10 000, and the most important Apply Masks to Tie Points),
- then run Build Mesh operation using new (experimental) meshing option using Medium quality depth maps.

Can you please try to reproduce the same workflow on our side and let me know, if the approach works? If so, then the result could be improved by improving the quality of the input images, for example, but using the image frame space more effectively, since currently the banana occupies less than 5% of the image image space, so from 13 MPix less than 1 MPix is actually used due to the applied mask.

[shantu2018]

I took a total of four photos. But when they are aligned, they are always at a height. But when I shoot the pears, they can be well aligned.

[zagnimarco1]

Hello i have the problem that the  photos of banan example don't align, i had masked  the entire first  photos  but  don't works, this is the log:2018-07-29 21:15:52 Agisoft PhotoScan Standard Version: 1.4.3 build 6529 (64 bit)
Unlucky  i cannot post the log because it exceed the 2000 characters.

[zagnimarco1]

Hello i have the problem that the  photos of banan example don't align, i had masked  the entire first  photos  but  don't works, this is the log:2018-07-29 21:15:52 Agisoft PhotoScan Standard Version: 1.4.3 build 6529 (64 bit)
Unlucky  i cannot post the log because it exceed the 2000 characters.
May be the "unable to load openCl"  is the guilty?
I tried to disable  from 6121 to   6128 and  the camera over banana seem aligned
Hallo i think to have solved , the mask was inverted,i tought  agis have to see the point of the table, i was wrong.

[bassistas]

Does this method also apply when you take photos of an object on a turntable and with the camera changing positions on the z axis? or is this only for static backgrounds and only with the object changing positions?

[bestable]

Does this method also apply when you take photos of an object on a turntable and with the camera changing positions on the z axis? or is this only for static backgrounds and only with the object changing positions?

any notes on this question from developers?

[PolarNick]

Yes, it works. I scanned an object on turntable from three angles (i.e. three circles of photos) and for each angle I just additionally took photo without object (i.e. photo of clean turntable + background). Then I masked these three photos fully and applied masks to tie points.