new to agisoft, turntable workflow?

[thebeals]

Hey all, new to the forum and to this software, I am trying to create a 3d shoe from a turntable.

My process:

I am using a lightbox that is designed for this sort of thing with a motorized turntable so it is well lit with soft lighting.
I have taken 16 pictures on a 360 degree rotation at 2 different locked camera angles to equal 32 photos total.
I mask all the photos.
I bring it into agisoft and apply the mask and align photos and build a dense point cloud

When I do this, it is only giving me half of the object that was photographed. It is only using photos from one side of the object.

5 camera angles from each angle are being used instead of the 32 recorded.


Why am I only getting half of the object? The detail is very good, but it's only half of the object. The object is a shoe that is similar in construction on each side. So I'm wondering if it is getting confused and thinking the two sides are really one.

Do you guys think this is the case? How do I force it to recognize the other side? I can't share the assets or show the shoe, but the attached image is a top down view of what I am seeing.

Thanks for any and all help!

[MartinBim]

Did you tried more overlapping photos? I'm using about 80 photos per turn.

[Alexey Pasumansky]

Hello thebeals,

Recently I have seen quite infomative video tutorial regarding turntable image capturing scenario:
https://www.youtube.com/watch?v=9_F-b2hxP_o

[featherdotcom]

Has the issue of all of the photos appearing to be on one side and only half of the object stitching been resolved? We are having the same problem and are trying to circumnavigate it by writing and inserting our own csv file, which places the cameras where we want them to be but then when we try to move forward these things happen:
1. Try to align and it reverts back to where the program perceives the angles to be and not what we wrote in. 
2. Try to skip align and use the update toolbar instead it gives us a "not enough reference data" pop up.

As we are working with fragile museum collection objects, using a turntable is ideal but we have been problem solving this stitching prob for awhile now to no avail.

Any advice/help appreciated!

[Alexey Pasumansky]

Hello featherdotcom,

There are several reasons that may cause incorrect or unstable alignment results for turn-table solution:

1) Not using masks to cover the static background. Using masks is obligatory for such method of the image acquisition when the camera position is fixed and the object is being rotated in front of it.

2) Lack of the texture pattern on the objects being scanned. For example, there will be likely reconstruction issues with the plain and untextured metal or plastic surfaces, shiny and glancing materials. Also mirror and transparent surfaces are almost impossible for reconstruction.

3) Image quality is not good (high noise, blur and etc.), low overlap, ineffective use of the image frame space can be also factors causing alignment issues.

Maybe you can provide some additional details about the type of object you have problems with, maybe a couple of screenshots or  original images preview?

[Melissa]

Here are some screenshots of our issues..

[FLuca]

Turntable works fine for me!
See below examples made with one camera and one flash unit:
https://skfb.ly/Gr9S
https://skfb.ly/FGWE
(both are not raw models and have been manually retopologised and cleaned)
Looking at your screen capture, it seems that the lighting is not consistent between pictures. Some are very bright, others are darker.
My flash unit is placed behind the camera and the light directed to the background and roof which are both white. This way I have soft lighting on the face of the object I am shooting. The camera shall be in manual mode to have consistent pictures.
It is hard to see on the screen capture, but it seems that your object surface does not have a lot of details except for the overall shape and is very shiny. As Alexey says, photogrammetry is not compatible with reflective objects, try polarised filter to reduce specularity.
In addition, try not having the full object in each photo, but make turns on the top, the middle and the bottom or even more, each time with at least 30% overlapping.
The advantage of having an object and not a life model, it that you can have as many pictures as you want 
Good luck
Fred

[scr1tch]

Hello. I found this thread when researching photogrammetry with objects on turntables. Some good tips have already been given here.
@ Fluca: Your examples look really great! Can you possibly give some details how you hated the photos taken, how many have you used and what you had to do on the PC nor to do any preliminary work to get these models like that? It would be nice if the whole thing is described as a beginner-friendly, then many other people who are dealing with this method just new, something of it

Maybe you can upload your raw image dataset here so that people can first try on perfectly captured images before they start with their own maybe not quite perfect images and stumble over pitfalls, which might also depend on the image quality. That would be great!