Fusion settings combinations for best detail

[KNSCI]

Hello;
Getting started with this SLS-1 and David 3.10 software.
We are trying to get the best detail we can in a textured, sculpture object about 40mm in height, with close valleys and ridges, with up to about 3mm depth in the valleys.
I scanned from high and low angles, cleaned each of 48 scans (24 hi, 24 lo; 15 deg increments), and with WalterMo's suggested alignment/fuse technique got some pretty amazing results.
Initially I did Fusions of the fully aligned scans at 250, 500, 700, and 1000 resolution settings, with full sized scans, 50% and 25% "Reduce Mesh Density" applied, no refinement or closing. Results ranged from smooth to very detailed with some anomalies in the 1000 Resolution fusions.
File sizes of resulting obj's appears related to Resolution setting, not to the "Reduce Mesh Density" settings, that is, file sizes tended to match within 1MB no matter full size, 50% or 25% density. Based on the screen shots, the level of detail didn't seem to vary much with density, more so with fusion resolution.
So I'm wondering what the Reduce Mesh Density does, I thought it was maybe akin to jpg compression settings in 2d photography. Visually it doesn't seem to do much.
Also wondering about the language of the command, is it Reduce Mesh Density TO x% of original scan, or BY x%? Would make a difference for all but the 50% setting.
Our CAD people, who'll be working in Rhino and ZBrush to create stls for 3d printing, want smaller files while retaining sufficient detail to do what they do and I'm trying to refine the settings/workflow I should be doing. We're also all challenged by widely varying internet speeds and hardware capabilities.
Finally, am wondering about the Smooth Scans, Shrink Scan Border, and Delaunay Optimization options for the scans, what those do and in what situation to use them.
Wow that's a lot of questions! Not sure the Subject line captures it all.

[ShadowTail]

Reduce Mesh Density applies to the selected Scanned Meshes and reduces them by the percentage represented as far as I can remember.

Fusion itself calculates an entirely NEW mesh based on the settings and scanned meshes.
A setting of 250 will generate a relatively low detail mesh.
A setting of 1000 will generate a finely detailed mesh.
A setting of 2000 will generate an insanely detailed mesh.

But its all based on the scanned meshes. If you reduce the details of those you will ultimately get less detail in the fusion mesh. Even at high settings.

Your CAD People should actually demand as high a resolution mesh as you can give them - they can always go and reduce the polycount using software like Meshlab or whatever options their tools have availale for that and go from there.

In general, for garden gnomes and such, a fusion detail level of 200-300 is likely more than sufficient while for coins, teeth and such objects you may want to crank up the detail to 1000+.

Hope this helps

[KNSCI]

Thanks, ShadowTail;
Good detailed easy-to-understand reply that helps quite a bit for this newb.

Knowing if it's Reduce TO or Reduce BY will help me choose the right settings when needed.

So we're saying it's kind of analogous to 2d photography, take a raw image at full resolution, work on that to retain the most detail, then reduce for other uses.

Ideally, we take the full scans, clean and align, then if hardware supports it, Fuse at full mesh reducing file size with the Fusion resolution setting per the end user needs, do I have that about right?

The weird thing is, whether I Fused with full size or reduced scans, the Fusion result was always about the same file size, so in this case of trying to make a workable size obj, the scan reduction didn't seem to have an effect.

Our CAD guy's laptop had only 4GB RAM and not a high end graphics card, so that was a limiting factor in feeding him fused obj's to work with. He requested max of 100MB files which were produced with a 700 Fusion Resolution, which was a compromise for his hardware--it was a file he could work with but lacked so much detail that it caused him more work in ZBrush before sending us the stl for printing. Now he's got 8GB RAM installed and some other hardware/software stuff going on so hopefully his machine can take the 1000 or higher Fusions, cause him less rework time, and get us all on a faster road to good print results.

Any idea or reference for what Smooth Scans, Shrink Scan Border, and Delaunay Optimization options do?

Many Thanks!

[ShadowTail]

The fusion result will be about the same size because the fusion resolution does not define the amount of detail you get but the actual density of the fusion mesh.
Higher density automagically results in more details being retained - unless you reduce the scanned meshes to a point where details are simply gone foerver.

Smooth scans will essentially remove "noise" from the scanned mesh and also reduce some of the fine details.

Shrink Scan Border removes the outer mesh points of a scanned mesh as those sometimes tend to be inaccurate (really depends on the scanned object and whichever way the surfaces face, too).
This usually removes stray polygons that are not connected to the main mesh as well as spikes and thus improves the alignment process.
Its sort of a simple cleanup process.

Delaunay Optimization essentially rebuilds/reorders the polygons of the scanned mesh into a more optimal configuration.
More detail here: http://www-ma2.upc.edu/rsilveira/pubs/F ... n_CGTA.pdf (long, complex article)

[hunkatibor2]

Hi
They asked me a question. STL format should be scanned to convert existing T-spline format. Can anyone help me how to do it?
Tibor

[KNSCI]

Shadowtail, thanks for the helpful follow up.

Interesting how the density works. It's like you're telling how many boxes of what size to put in there and no matter what's in there the size of the file will be the same. So better to not Reduce Scans if interested in saving file size. Provided the hardware can handle it, use the full scans then.

Today I did a series of fusions after improving some scan alignments and saved as .stl's. B9 Creator was not happy with the nearly 200MB 1000 Resolution stls, crashed every time I added one. Will have to do my print experiment with lower resolutions. Want to put 8 little models on the table from 250 to 1000 resolution, one series of full resolution scans fused, and the other with smoothing applied before fusing, so will probably just do the 6 from 250-700.

I'll have to try Shrink Scan Border too, I spent a fair bit of time Cleaning up islands, extraneous bits, and sawtooth edges off the aligned scans in the hope of making a nicer Fusion. Sounds like that option would help. When is a good time to apply it, during post-scan cleanup or after alignment just before fusing?

I'd asked about the Delaunay since the google results I got were either papers you need to pay for or were over my head from the abstract already! I'll take a look at the link you so kindly provided, see if I can determine if the more optimal configuration would help in what I'm doing now.

Many Thanks!

[hunkatibor2]

Hi KNCSI
I would like to see your work, send us pictures.
Tibor

[KNSCI]

[Update]
the six laboriously scanned, cleaned, aligned, smoothed, fused and saved stls, carefully added to B9 Layout and printed, showed almost no difference in surface texture detail although on screen the 250 resolution fusions looked quite a lot smoother than the 700 or 1000's, and the smoothed scans provide less on-screen detail that the unsmoothed ones.
Could be that we had the table set for 75 µm xy resolution and sliced at 76.2, but at 3 hours print time that we didn't have the time to do anything finer.
Might try one set next time we move the table up to a higher xy and I have more time to print, or another object of similar size.
Although a pretty fair likeness, kind of disappointing so far, as I'd hoped to be able to print the texture detail as well as it looked on screen.
That's the learnin' process I suppose!

[Artisan S]

Delauney optimisation......a method to represent a surface by retriangulating a pointset by taking the crosssectional points of circles starting from the original point. The purpose of a Delaunay optimisation (btw Delaunay is a Frenchification of a Russian name) is to reduce (drastically) the number of tight corners in a triangulated mesh which makes texturing a little more stable (textures don't like tight (< 30 degrees) corners in a mesh. Therefore softbodies are usually squared (no tight corners), hard bodies can use triangles but beware of sharp pointy corners.

So now we have that cleared up lets get to the problem of the surface detail. I've got "some" experience with 3d printing (having used FDM and SLA printers up to a Project HD 3000 plus). Now you are using a B9 creator right (great choise btw, I have one firmly planted on my whishlist, if I don't build an SLA myself). Preserving surface detail depends on a lot of things in the proces from scan to print.....things to consider:

1) How deep do I scan in order to find detail......detail in the scan is determined by three parameters

- The resolution of the scan camera (this determines the maximum resulution since the algoritm determins the corners of each pixel and checks the xyz coordinates for all of them) but remember that is a compound of:
a) the lens (the better the more detail will be picked up) and the accuracy of focus
b) the aperture (smaller aperture means more DOF means a more consistent sharpness througout the scan, which in turn means better allignment and that results in more resolution)
c) the noise (since noise in a 3d scanner camera is translated into surface wobble and wobble is bad in two ways
- wobble reduces detail direct (duuuh)
- wobble reduces detail indirect (since allignment will be less accurate)
d) the artifacts of the compression algorithm (that's probably the reason David modifies the drivers of their own cameras)
- The resolution of the beamer (now since the camera determins the actual points of the mesh the resolution of the beamer is less critical in my experience) but there are a few things to consider:
a) actual detail is picked up not by the sharp projected patterns but by the out of focus patters.....I know this is counterintuitive (I was flabberghasted when I found out but hey who am I) but it's in the algorithems.
b) in order to enhance detail it pays to set the beamer just a tad unsharp thus reducing noise from the beamer
- The quality of the allignment and this can be flunked by:
a) crappy calibration so high quality calibration corners (or better known quality calibration corners since corrections can be made prior to calibration in the advanced settings) are essential
b) the quality settings of the scan (to loose and crap triangles ruin the alligning to tight and you have lack of overlap between the scans)
c) the technique of alligning (sometimes I realign scans using a different way of alligning and achieve much improved results) since I scan static objects (mostly) I use a Arduino controled rotor that is 1/400 preciese and allign on Y-axis

Now that is cleared as well but you are happy about the surface detail in the scan, so this has been done right (righ?).

Now to the next step printing. There are a few things to consider here as well. Indeed even using a SLA printer like the B9 creator you need to consider the detail in the scan. The fact that a printer can reach a resolution of 0.1 mm does not spill over in the fact that the smallest printable detail is 0.01 mm ass well since a detail is created from several points spaced 0.1 mm appart. Now the resolution of the B9 is 50 microns which is 5/100 of a mm. But the build erea get smaller the more resolution you are using so that depends on the size of the overall print. But 50 microns is impressive (not when you are used to professional printers like a HD 3000 plus but never the less for the price the B9 ROCKS). But is that smooth. Not quite, there will be some noise in the print reducing the perceived detail to about 0.1 of a mm and if you surface detail is smaller then that you will be prone to losing it in the process. David can scan it (I've had scans a crazy resolutions, using a a modified webcam (2mm objects are described on the forum, thats even possible) but do not expect that from any printer). Also the slicer settings are important......I've checked into the manual and I have found out that some resin is more detail freindly then others.....RED for instance is triggered for 100 micron (that is 0.1 of a mm) but not for 50 micron.

Greats Ed.

[KNSCI]

Artisan S;
Many thanks for the detailed reply!
I have to do some searching here for using David projector with Canon 7D, see if that helps. Our project that needed the scanner and B9 for fine resolution reproductions went away so actually have been busy using B9 for potential customer sample prints, and the David not so much.
Dave