Hi,
I’m trying to understand which unwrap style fits best for my use case. However I’m not able to tell much difference, visually, among the three styles: maximal textures count, fixed texel size and adaptive texel size.
They all generate very tiny texture components and leave 50% or more of the image real estate unused. Moreover, the resulting texture is close to impossible to postprocess with photoshop, or similar application.
Is there any way to generate larger texture components, which would use more effectively the image space ? Any other unwrap style which I may be missing for some reason ?
thanks
Massimo
Hi Massimo Tasso
Use the FIXED TEXEL SIZE and use the OPTIMAL TEXEL VALUE in the UNWRAP tool. It depends on the use case but I would use 4k-8k textures as the maximum texture resolution…
Wishgranter wrote:
Hi Massimo Tasso
Use the FIXED TEXEL SIZE and use the OPTIMAL TEXEL VALUE in the UNWRAP tool. It depends on the use case but I would use 4k-8k textures as the maximum texture resolution…
thank you. I’m starting to understand how it works, and making a few tests. The optimal texel value is still not optimal for my models, I achieve better results with half of it or even less than half. The texture image real estate space is used much better with lower values, while teh optimal texel size fills it no more than 25/30%
Could this be because I select a small portion of the scene as the reconstruction region ?
Hi Massimo Tasso
Yes, as using it on small area it can create another optimal texel values… Watch TEXTURE QUALITY number… It needs to be 100% and image downscale set to 1 before texturing to get the best results regarding texture resolution…
Wishgranter wrote:
Hi Massimo Tasso
Yes, as using it on small area it can create another optimal texel values… Watch TEXTURE QUALITY number… It needs to be 100% and image downscale set to 1 before texturing to get the best results regarding texture resolution…
thanks for the tips. Where do I find the TEXTURE QUALITY number ?
I’m now testing the texture creation in the photo consistency based style. The first try did not work (PC got stuck), hope that leads to good results.
I guess you guys have to work a bit harder on the texturing algorithm, because as it is today it is not on par with the rest of the reconstruction process (which is excellent and extremely fast) and not with the competition. Unfortynately a bad texture can spoil the results of the entire workflow, at least for my use case, because I can’t deliver figurines with poor textures.
If the registration export to bundler was working properly (it was not, the last time I tried some 10 days ago) it would be possible for me to bypass this issue.
Hi Massimo
I’m now testing the texture creation in the photo consistency based style. The first try did not work (PC got stuck), hope that leads to good results.
I highly recommend you to use the VISIBILITY-based option…
thanks for the tips. Where do I find the TEXTURE QUALITY number ?
Just click on the UVed model and get all information… look at the screenshot…
Model_info.jpg
I guess you guys have to work a bit harder on the texturing algorithm, because as it is today it is not on par with the rest of the reconstruction process (which is excellent and extremely fast) and not with the competition. Unfortynately a bad texture can spoil the results of the entire workflow, at least for my use case, because I can’t deliver figurines with poor textures.
You need to dig a bit deeper and do some testing on a smaller dataset ( say 50 images is OK ) with various settings so that you can see their impact on the results…
If you set it up properly, you can get very GOOD results…
thanks, found it. For the example model I’m using to experiment it shows texture quality 100% and texture utilization 16% 
How do you embed images in the posts ?
Hi Massimo
thanks, found it. For the example model I’m using to experiment it shows texture quality 100% and texture utilization 16%
Now just lower the TEXTURE RESOLUTION from 16k (??) to say 8k and get higher texture utilisation as now it uses 16% from 16k (??) texture space… It all can be set in the UNWRAP tool…
Regarding uploads:
Screenshot 2016-12-06 17.08.46.png
Wishgranter wrote:
Hi Massimo Tasso
thanks, found it. For the example model I’m using to experiment it shows texture quality 100% and texture utilization 16%
now just lower TEXTURE RESOLUTION from 16k ?? to say 8k and get higher texture utilisation as now it uses 16% from 16k?? txtr space… al can be set in UNWRAP tool…
I always used 8K
Hi Massimo
then it used only 16% from the 8k map ( I was not sure what resolution were you using 
)
So set it to 4k and use UNWRAP…
Hi Wishgranter
Wishgranter wrote:
you need dig a bit deeper and do testing on smaller dataset ( say 50 image is OK ) various settings so you can see their impact on results…
If you setup it properly can get very GOOD results…
honestly, I doubt that just by setting it up properly I can achieve results like this
texture-ps.png
when all I get, after several experiments, is something like this
texture-RC.png
. Other parts of the texture are otherwise OK.
Unfortunately I cannot sell my products in the latter case. Therefore, I definetly need a patch to do a proper export to bundler and be able to produce the first type of texture. Otherwise I will just need to fall back to the old, slow but good quality tool and workflow. It would be a real pity, both for me and for you guys. Apart from this, Reality Capture is really one meter above the competition, but this one takes it one meter below, at least for my use case.
Maybe you could ask the developer to tip me in fixing the bundler.out creation ?
what I get by importing bundler.out is the cylinder of cameras perfectly aligned and the reconstructed mesh normal to it rather than coaxial.
if I export bundler with negative z, I get the cylinder of cameras perfectly aligned, the reconstructed mesh coaxial with it but translated along the z-axis of a full length of the mesh itself.
Did some experiments in swapping the variables signs in calibration.xml, but I can’t get a sense of it.
thanks in advance
Hi Massimo Tasso
Your example looks like misalignment of cameras, what sort of subject are you scanning there ? can you post an overall view on the subject ?
Wishgranter wrote:
Hi Massimo Tasso
THe example there look like misalignment of cameras, what sort of subject are you scanning there ? can you post overall view on the subject ?
Sure, I’ll post it tomorrow morning. Indeed the cameras are aligned, the mesh is especially clean and neat right on that spot. Anyway, it’s a person and the detail is his hat.
here is the full face, I can’t show the whole figure. Please disregard the awful colors, I used meshlab which is an awful visualiser. Colors are OK.
I’ve circled the areas where the texture fails, basically everywhere there’s fine details.
face.jpg
Hi Massimo Tasso
Why don’t you use the RENDER tool to get the original colors ? You do not need to use Meshlab or anyhing else
Apparently it’s an ALIGNMENT issue… are you using RPI cameras or DSLRs ?
Can you send me the dataset for inspection ?
Wishgranter wrote:
Hi Massimo Tasso
Why don’t you use the RENDER tool to get the original colors ? You do not need to use Meshlab or anyhing else 
Apparently it’s an ALIGNMENT issue… are you using RPI cameras or DSLRs ?
Can you send me the dataset for inspection ?
I’ve generated the texture and saved the the textured mesh. My version of RC unfortunately does not show yet the closed model but only the points, therefore if you want a screenshot I need to open another application, meshlab in this case.
I could have opened it with 3dCoat and the colors would have been OK, but 3Dcoat is not installed on the PC I was using when I took the screenshots.
Yes, I’m using raspberries, and this affects the mesh detail quality with other programs, but not with RC (far superior in rendering the shape). Unfortunately the textures come out like this, and it’s not a problem of photo quality (8 Megapixels) because with other programs the mesh may be mid quality but the texture is perfect and crisp, starting from the same pictures.
Hi Massimo Tasso
as I’ve already said, look at the RECONSTRUCTION->RENDER tool and use it, no need to jump out of RC
And, as I have asked, is it possible to get the data for inspection - RCPROJ file + folder + images used here ?
Wishgranter wrote:
Hi Massimo Tasso
as i say previously look in RECONSTRUCTION->RENDER tool and use it.no need to jump out of RC
And as have asped its possible to get the data for inspection ? RCPROJ file + folder + images used here ?
I have no “render tool” in the reconstruction tab.
rendertool.JPG
Maybe I need to upgrade, but how ?
I will check if I can send the project files and photos. Thanks for the offer to look into this
Hi Massimo Tasso
So what is this then?
Screenshot 2016-12-07 16.40.13.png
oh, I didn’t get you meant that .
Last time I tried it it didn’t allow me to select what parts of the model to shoot, so I just forgot about it