Problems with relatively thin objects

[Opparg]

Hello!
I keep running into problems with relatively thin (not completely) objects, such as this slice of bread. Sometimes they work, sometimes they don't.  I have photographed it by rotating it in front of the camera, then from 45 degrees above, and then flipping the object, doing it again. Sometimes it works, sometimes I get results like the first screenshot.

As shown in the second picture, the cameras are all over the place.   I had -some- better luck with photographing the object standing up instead of laying down, but then it refused to align the pictures taken from the front, and only used the ones from 45 degrees above.

The pictures are taken in a light tent with a Canon 70D  If I photograph a thicker object, like a whole loaf of bread, it works perfectly, 100% of the time. Same thing with rocks..  Cylindrical objects work 50-70% of the time if they are laying down, if I do them standing up I get more problems..

The third picture shows the series of pictures I have.  I can't figure out where I can improve it. Is it in the photos, or the alignment?  All of the pictures are masked.. I could really use some help in straightening out this process, thank you!

[Kjellis85]

Try not moving the object that much between each frame. Also, you could try setting the angle a bit lower  so that you get more of an oblique view. This should improve alignment.

[Opparg]

Ahh so parallax has a significant effect on the scanning result?  I heard it mentioned somewhere but didn't realize it was that important.  Maybe I should find a turn table. Would that help?

[bigben]

Fill the frame more, and possibly try linear passes along the bread rather than circling around it. I find this works more reliably for longer objects.

[Opparg]

Linear passes.. interesting. So how many different angles do I do that with?   Say I have it from the front, do a linear pass, rotate it 45 degrees, do another, etc?

[Alexey Pasumansky]

Hello Opparg,

Maybe you can send us the original data and the project file to support@agisoft.com?

[Opparg]

Absolutely. How do I send it though, since it's 300mb with the photos? (I thought that's usually too big for emails)

Thanks!

[gatsri]

check www.wetransfer.com
there you can send up to 2 GB free, the client will recive a link to download the dataset.

[Alexey Pasumansky]

Hello Opparg,

Thank you for sending the data.

Looks like the problem is in the lack of overlap between bottom/top images and side sub-sets. I think that the better way is to reshoot the object with more overlap between images made on the different height.
Also I can suggest to reduce the ISO.

[Opparg]

I will add another angle for the rotation and see if that works.  And yeah the ISO is high, I had an extreme f-stop, as high as 28 sometimes to get more focal depth, but from what I have seen from others, this is not necessary, I could go down to somewhere between 8-14.

What you say makes sense, it's as if the alignment needs more data to connect the different heights. I will try this today and see how it works. Thanks for the fast response!

[Alexey Pasumansky]

Also, as mentioned above, you can use frame space more effectively.

[Opparg]

Getting much better results now. The biggest issue was actually the light. More frontal light, and taking advantage of daylight even with the light tent made a big difference.  The challenge is f-stop/iso ratio