Creating photoreal Alphas for Zbrush with Photosculpt

[hipe0]

Hi Zbrush users,

I've just posted this tutorial about my new unique method to create hyperrealistic alphas for use in Zbrush and Sculptris.

http://www.photosculpt.net/dump/2010...g-in-zbru.html

I think it's a really innovative solution as depth is based on analysing 2 photos instead of only one. This makes it of extremely high photorealistic quality. Who's interrested to try?

Thanks

Hippolyte from www.photosculpt.net

[str13d3r]

I want to! It looks like something I always wanted to have. The results of other other Bump- and Normal Map Creators are very dissatisfying to me.

[CDevens]

The second step says to use Photoshop to add a blurred circle to mask the edges. The alpha palette has a radial fade function that does this for you, so you don't need that step and you can just use the image that was made in Photosculpt. Let me know if there is a reason for this step that I am missing, otherwise I hope this helps.

[hipe0]

100% right. Zbrush users can skip that step entirely because zbrush does a very good job with the radial fade feature.

For Sculptris users or users of other software this step is needed to avoid hard edges at the border of your tool

[hipe0]

I want to! It looks like something I always wanted to have. The results of other other Bump- and Normal Map Creators are very dissatisfying to me.

I think I can help.

Photosculpt software uses 2 photos of a subject and combines them with an in-house-made (by-me) binocular vision algorithm.

The direct result is a huge step-up in quality for depth maps and normal maps. The difference is just amazing.
I know exactly what you mean by "other softwares very dissatisfying" but Photosculpt cannot be compared with softwares that use only 1 photo.

Think of it more like a 2.5D scanner, not quite full 3D yet I'm working on this, but extremely capable already, really easy and fun to use and results are high-res, fast and naturally photoreal.

Photosculpt v1 is for sale since march 2010 now with very high reviews. You can now download the 30 day trial on my website and potentially scan anything 2.5D from your surrounding into a hi res Zbrush alpha tool within minutes.
Creative possibilities are endless.

I highly recommend Photosculpt for any Zbrush artist with access to a camera.

Don't hesitate to comment. As the creator of the software, your feedback is warmly welcomed as I could be bringing new features for better integration into Zbrush very quickly.

[Rastaman]

Detailed 3D models from just 2 simple fotos in 20° angle ?
No need for an expensive 3d-scan-system ?
Sounds cool so far.

Hmm, i tried it with some shots, also lined them up and corrected them with photoshop before use in the demo, but my results are very poor, not comparable in any way with your demos on the web-site. All attempts resulted in somehow distorted meshes.
Tried it also with a face shot, but also no suitable results for use as detailed maps for characters.
Results look more like a crumbled sheet of paper than a stereo-model or scan like the ones Lee Smith generates with his tripleganger-scans.


first 2 pics are the shots, second two the resulted mesh from 2 angles.


And another test with a closeup, no warnings from the program, but still a crumpled mesh:





And a last try under stable light:



No luck. Seems the program doesn't like my photos.

[hipe0]

Hi rastaman,

the software doesn't like your subject choice.
Thanks a lot for actually posting your results. I see that you did it very seriously.
Your experience can be helpfull for anyone who starts with any photogrammetry software.

your problem is : Not all subjects are best for photogrammetry.

To explain this I need to explain a bit of how a photogrammetry software like Photosculpt works. It can't hurt as other softwares basically also work the same.

First the software tries to match all pixels in patches from left photo to another patch of pixel from right photo.
Second, the software does parallax computation: most patches will move along x (some along y too) with different amounts. Displacement along x is an excellent givaway about depth of the pixel.
Then it's the easiest part, the software just encode the depth map or normal map or mesh using calculated depth.


But understanding this you'll see a few problems.

1. uniform color objects without textures: not optimal. a lot of patches will match the uniform color. resulting in either increased noise or worse: artifacts like the one you showed.
2. translucent objects: won't work
3. mirroring objects like cars: won't work
4. complex multiple depth objects like tree branches: won't work
5. white objects: not optimal
6. black objects: not optimal
7. lighting inconsistent: not optimal
8. photo blurred: not optimal

The optimum subject for Photosculpt are not too deep, highly textured, with uniform background and soft outside lighting without flash.

I'm going thru this again in my Tutorial #2 (5 minutes)
http://www.photosculpt.net/dump/2010...m-results.html

In the tutorial I recommend to start shooting first a simple subject, a stone wall.
And go to more demanding subjects once you gain experience about camera angles, lighting.

Amateur photographers can then read the advanced tutorial 6 here (5 minutes) to optimize shooting
http://www.photosculpt.net/dump/2010...ptimum-re.html

So again, I'm sorry it didn't worked for you the first time. But don't give up, once you get used, it's really fun!
By the way I've opened a new gallery this week http://www.photosculpt.net/gallery with fantastic 2048x2048 seamless tileable textures + all maps for free download, of course created with my photosculpt software.

Best

Detailed 3D models from just 2 simple fotos in 20� angle ?
No need for an expensive 3d-scan-system ?
Sounds cool so far.

Hmm, i tried it with some shots, also lined them up and corrected them with photoshop before use in the demo, but my results are very poor, not comparable in any way with your demos on the web-site. All attempts resulted in somehow distorted meshes.
Tried it also with a face shot, but also no suitable results for use as detailed maps for characters.
Results look more like a crumbled sheet of paper than a stereo-model or scan like the ones Lee Smith generates with his tripleganger-scans.


first 2 pics are the shots, second two the resulted mesh from 2 angles.


And another test with a closeup, no warnings from the program, but still a crumpled mesh:

And a last try under stable light:

No luck. Seems the program doesn't like my photos.

[Rastaman]

All right, I'll try other samples.
Thanks for the explanation.
Would be cool if you could evolve it further so it could be used for human skin and faces / bodys.