model creation faild

[Anonymous]

this is my third project with same problem
it seem like i did something wrong while taking photos
so i need to know what i did wrong ................

i attached sample photos and a screenshot of the model 

sorry for my bad English .....
 

[tehj · Accepted solution]

Hi, 

Can I ask how many photos you took?

Also just to confirm, it looks like you are placing the object on a table and turning it by hand; is that so? 

The reconstructed mesh looks like the engine was not able to identify the camera positions from the photos, which could be due to drastic changes of the same object, or very strong similarities in different photos (e.g. the object moves in a still background, or repetitive patterns) 

In this case it looks like the background was identified as the same in every photo, while the object was moved. This will cause the reconstruction engine to fail to properly predict the camera positions relative to the object (it still thinks it is looking at the object from the same side even after it was turned). 

My suggestion is to either use a featureless background (e.g. plain colored wall or surface), so that the object becomes the main point of focus for the engine, or to take pictures by moving around the object rather than by turning it. 

Cheers,

Julian

[Anonymous]

thank you for your help,

I took about 80 photos from 2 levels...
and yes i am placing the object on a table and turning it by hand with constant background.

I did what you say and the result was good not enough but good.
I took 114 photos from 3 levels

[tehj]

Hi,

I'm glad to hear that the results have improved! 

It looks like the top part of your object was not recognized enough. This can happen for a few reasons, such as insufficient photos, drastic changes across photos (more or less the same as before), or because the object itself was too featureless for the engine to correctly identify. 

If what you want is the mesh, then I would suggest sticking some markers onto the top (multi-colored stickers would do, although it is also possible to get markers that work well with computer vision techniques). If you want the texture on top of the mesh, then I would suggest just taking more photos of the top of the object from different angles (which is a somewhat inefficient way to increase the number of features).

Cheers,

Julian