I am taking pictures of people’s heads to then create a 3D model from it.
However since people sometimes slightly move etc. i get sometimes bad results (which isn’t all that bad since i can use symmetry in 3ds Max).
The only thing that is bugging me is that when I import photos and press allignment, i get different results with the same pictures.
For example at the first try I only get the back of someone’s head, and on the second try i get their face and neck.
Why is this happening and how can I resolve this, I am new to the software so i’m still a noob.
Hi Jerremia,
I thought the asymmetry of peoples faces contribute enormously to the indiviual expression? 
Anyway, what you experience is the result of a randomness somewhere in the pipeline of the algorythms. If the image set is as perfect as it can be, the effects are almost not noticable, only in figures 2 or 3 after the point in mean or median error. However, when they are not ideal and at the border of breaking up (or like in your case obviously beyond), this effect is very pronounced. My guess is that this randomness is part of the secret of the RCs speed, but I might be entirely wrong. Take it as an indicator for the suitability of your images to be processed by RC.
Happens to everybody who first uses RC and is not a trained expert…
When there is more than one possible solution for a given alignment, RC will pick one at random.
you’re right asymmetry is very important when it comes to people’s faces. The project for which I need the 3d models, is based for AR and they’re relatively small. And on top of that it’s more of a gimmick app for christmass rather than something i’m doing for a client who’s paying 
What do you exactly mean with :“only in figures 2 or 3 after the point in mean or median error. However, when they are not ideal and at the border of breaking up…”?
We used an actual own made 3D scanner last year, which 100% replicates humans (and faces). But this scanner was relocated to a different country for business reasons.
This is why we’ve resorted to use a more “rough” technique, taking pictures with a canon camera etc. and combining it in RC. I’ve had atleast 1 succesfull scann of someone’s entire head.
I think for this technique that i’m using, i have to take atleast 60-120 pictures and make sure the person who i’m making pictures off is not being distracted by other collegues and stays still during the entire process.
Thank you for your time!
@ShadowTail Thank you for the information, this probably is the reason why i’m getting faces inside of faces.
Hi Jerremia,
no problem!
The faces inside of faces can also be an alignment problem, as in that there are two independent groups of images and each has their own idea of where the face should be. This is the reason for pockmarked surfaces.
There is literally no way of preventing movement over an extended period of time, that means only with a multi camera rig you can be guarranteed a successful alignment of all images. This has been discussed here a few times already.
What I meant is that the differences for the mean and median errors (both in the alignment report) will only differ 2 or 3 places behind the separator, as in fractional. Sorry, my mathematical english is a bit rusty…
And by breaking up I mean that you get several components as opposed to just one. With experience and a still object, it’s no problem to get 99.9-100% at the first try. Just had 1191 out of 1192 images, and one was only an accidental shot and ultra blurry…
Thanks again for the reply!
This information is very usefull, we do have a multy camera setup for small objects that another collegue is working on.
But if I want to use that i’ll have to chop people’s heads off, and that’s not a very good solution I think.
Again thank you for your time, and if we get serious with humans again. I’ll be sure to either get another “copy” of our existing 3D scanner or create a rig as you mentioned.