Texture resolution error on export

[Flow]

Hello everybody.
I've got a strange and enoying error with textures size on export.
I have got a 3D head to reconstruct in high resolution and need at least 8K texture map at the end (maybe more because the texture will be reproject on a clean model later in another software).

The matter is that, whatever I do, Zephyr export small (and random size) textures.


My parameters for Texture generation :


My parameters for export :


-The resulting texture size in this case is 3072x3072 ?!


Sometime it's a bit more ... or less.

And if I rescale texture on export, instead of "none" option, the soft just rescale the texture on export as it says,
That mean that I have just a 3072x3072 upscaled in 8K.
Logical, but I can't see any solution to solve my problem.

Any idea ?

[cam3d]

What kind of cameras are you using for the reconstruction? If they are low MPX like rPi cams then it may simply be that there is not enough resolution in the data-set to provide a full 8k texture.

The 'randomness' is probably a result of slightly different uv packing - approximations are made to speed up processes and this can lead to slightly different results. If the difference varies A LOT then it might be cause for concern, but otherwise it's perfectly normal.

[Flow]

I don't really understand your answer.
It's a professionnel studio rig with 76 cannon 1200D (I used43 for this purpose), and 5184x3456 px photos. I even try to upscale them (x2 and x4) with an AI upscale software, but nothing work.

The fact is if I ask 8K textures, the software should provide 8K textures and not deciding himself wich resolution he want to offer me, isnt it ?

And UV's never had influence on texture res in any software until today, so I don't understand how it can modify the output res ?

[cam3d]

Is it possible to share the data-set? Without seeing it first hand it's hard to figure out what the problem might be. Feel free to send it through to support@3dflow.net rather than through the forum and I can contact you privately through there.

Just because you ask for 8k textures doesn't necessarily mean that there is enough pixel data to provide a true 8k texture, especially if the subject takes up a small percentage of the frame. I've got a bit of experience with multi-camera systems (I have built a couple myself), so I understand some of unique workflow issues which arise with this sort of capture stage and want to help you as much as I can!

EDIT: The texture resolution you choose in Zephyr is the MAX resolution it will aim to provide you - For example if the head is only occupying 20% of the frame, you'll see much lower resolution results than if it's 80% of the frame. There is no point creating an 8K texture when the source data isn't high enough resolution.

[Roberto]

I want to add some notes to the discussion. Cam is right, the resolution value is a maximum value. Zephyr never goes under the one-pixel size during texture reprojection (as it would mean "upscaling" the original pictures). Usually, the software should also prefer the closest cameras with a good angle. However, if you have closeups photos of the skin that you think that were not considered for texturing in this case, try to make a new textured mesh subselecting only the closest cameras. If by any chance this solves your issue, then we should have a look at the dataset to understand why those cameras weren't considered. Thanks!

[Flow]

Thanks for your replys.
I Uderstand the concept of space occupyied and ever think about it.
I just had a doubt on the term "max texture size" and you answered it.
That's why I upscaled the picture before with a good AI algorithme to raise 20736x13824px.
But the result is still a small texture, and I don,t understand it.

As Roberto suggest it, selecting only the closer cam lead to a bigger texture : 5984x5984


I will dig on this side and let you know if I found a solution to have better results.

Thanks guys

[Flow]

Ok. Finaly, with X2 pictures and slecting some close up camera, I raise the so wanted 8K !
Here is the process :

1-upscaling pictures with "Topaze Giga Pixel" (or other) x2 size
(Unable to load x4 on my computer)
2-upscaling masks at the exact same scale in Photoshop or "Topaz"
3-decrease some parameter in Zephyr like matching camera precision and % of pictures used for meshing at 50
4-eventually selecting some closer cameras and reject the most far ones to get an bigger average size on the final texture

The result is prety clean and I obtain 8K sharp and detailed textures.


Indeed, in my case I need more definition on the face than on the rest of the body.
To do that, I shot more closer picture on the face than for the reste.
When I open these close up of the face in photoshop, I get approximatively 4k pixel just on the area the face.



A good thing will be to let the user choose him self between 2 options.


I know that Zephyr allow to choose a definition at export time, but this is a bit dark option because the user don't even know if the software recompte a bigger res or just take the res the algorythme foud (3k for exemple) and just upscale it. The truth is that I think the second theory is the good one, and this is a useless otption in that case.

[Andrea Alessi]

In case somebody is finding this thread in the future, please note that upscaling is not recommended as - see this thread here for more information https://www.3dflow.net/forums/forum/...-loading-issue