Problems with masking

[Snek]

[Andrea Alessi]

Hi Snek,

it's hard to tell exactly without taking a look at the photos.

From what you say, I suspect that your subject is not very textured and that makes it difficult for zephyr. In that case, you need to either take more pictures or use advanced techniques to help the SfM phase (e.g. a projector, or more simply, dirtying the surface).

If you can share the photos, I'll be happy to give you more specific advice.

[Snek]

Thank you I will try to use a subject with more texture.
I will post again if it won't help

[artsomnia.art]

Roberto and Andrea are the referent persons who can see deep into technical issues, but let me add something due to my personal experience:
First I have to add, that I'm aware of the fact, that Zephyr can reconstruct images made in any order, so you do not have to follow the ortodox photogrammetry image capturing method, by doing circles. That being said and it's works like that, but there are still some rules of photogrammetry
1)
- if you make bigger steps than 10 degrees till your capturing process it can end up that the software (any photogrammetry software) will not be able to recognize some cameras. Personally I found the 15 degree as a cap, so if you have 24 images in a full 360 circle around the object, it will probablly work out, but I would recoment 36 images till a 360 circle if you wanna go for perfection.
- you mentioned that you worked with 49 images, so basically it should work out, if you made 24 images in one circle, than 25 after you rotated the object, but if you made 2 circles befor rotating and another one after rotation, it could not do the math based on the 49 image cap. I olso have to add, that there are some complex objects that are imposible to handle by 2 circles with a rotation without heavy post reconstruction.

So try to make 2 clean circles as 24/25 images and see how it goes.

2)
- be sure, that you rotate the object well. If you rotateing it manualy it could end up as you mentioned, by lost camera positions, since it could happen that the software will recognize diferent images as panorama photos and not images that are covering the same shape from different view degrees. It's better to use turntable with greenbox (lightbox, black box, bluebox), or walk circles around the object, and generate separated background for masking with lights (object on a black homogenic surface, luminate it with spotlight, minimum from 4 directions and keep the room dark where you taking the pictures, so the camera will see, the base as black, the background as black since just the object is luminated on a black base)

Good luck

[cam3d]

Snek - when you say you took 49 pictures, do you mean screenshot or actual photograph?

Couple of other things I'd like to add - There is potential for the masks to be inverted, so might actually be stripping out the subject and trying to reconstruct from the background alone. In Masquerade RED is the non-masked, and blue is the masked (image attached for your reference).

The other thing was, 50 or less images is fine for feature rich subject matter, but if your subject is less than ideal for camera orientation, you can always demo one of the paid versions of Zephyr for 14days - Personally I take a minimum of 144 images for a full 360 capture, usually more, rarely less. https://www.3dflow.net/3df-zephyr-ev...download-page/ (Lite enables you to use up to 500)

[Snek]

I took Screenshots but with the suggsetet tip I managed to somehow make it. Thanks to everyone