Collecting tips for creating awesome 3D scans


(Bart) #1

I'd like to write a blogpost on how to create great looking 3D scans, and I'd like to collect your tips & tricks. Which software do you use? How do you prepare your subject for scanning, and how do you proceed from there? Any post-processing tips?

Please share your best practices here so our entire community can benefit.

Thanks!


(Discoveryprogramme) #2

Hi. A lot of our models are generated using laser scan data gathered from a faro focus. We find it is best to sample the data down before beginning any of the modelling stages, usually this sampling will be to 1cm or less, depending on the overall scale of the structure being surveyed. We then use geomagic to create a wrap over the pointcloud data. At this stage the wrap may runs into tens of millions of triangles, so we again sample the data down a few times to create different levels of surface wraps. These levels are then used to extract a normal map and ambiant occlusion which can be overlayed in sketchfab.


(Scansource3d) #3

There are numerous scanning technologies out there that all have advantages and disadvantages. We use Artec scanners to capture objects. Full disclosure: We do distribute for Artec. These scanners use a projected light pattern as opposed to a laser. One or more geometry cameras measure the distortions in the grid to measure the object and build your point cloud. Artec scanners have the added benefit of capturing color texture during this process.

Here is a quick video we did that gives very simple info about 3D scanning. It also briefly shows what the scanning process looks like with the Artec Eva:

Most of the smaller objects we upload to Sketchfab are scanned with an Artec Spider. Small meaning most items that would fit in a 12" cube. We use the Artec Eva for larger items with less detail. We'll be posting some of those at some point in the future.

Best practices for scanning and post processing with Artec are similar to other scanning technologies:

  • Shiny, highly reflective, or transparent objects can be difficult to scan. You can spray these objects with gray primer or, for a less permanent solution, dust the objects with talcum powder, corn starch, etc.
  • Very dark surfaces can also be difficult for laser and projected light scanners. Again, you have the option of dusting the surface and/or increasing scan sensitivity in the case of the Artec scanners.
  • Hair and fur is always difficult to scan due to the extremely fine details. Most scanning applications approximate the hair shape and product decent results. Again, increasing scan sensitivity with the Artec scanners helps with this.
  • Once we have the digital scan, we usually decimate the model. Many of the scans are millions of polygons. Most applications require much less. Reducing the polygon count will also reduce detail, though. The Artec software does a great job of reducing models. ZBrush also has an intelligent decimation tool called ZRemesher that does an incredible job of reducing a mesh while keeping the fine detail.
  • If you had to dust the surface or if your scanner does not capture color texture info, you can add material types and re-paint texture digitally if needed. ZBrush is great for sculpting geometry and painting or projecting texture. MARI is also a great dedicated digital 3D painting program.

We realize the Artec scanners are too expensive for most hobbyists just getting started in scanning. Photogrammetry is a very commonly used 3D capture method. You can use pretty much any camera to take a series of images and then use software to build your model.

Wikipedia has a great list of software options:

We have also had the opportunity to see the Fuel3D scanner action. It produces very highly detailed scans and is about $1500 USD. Fuel3D works similar to a camera in that you just press a button to create a capture. While it has a small field of view, you are able to use their included software to stitch multiple captures together to form a complete model of larger items.

Fuel3D
http://fuel-3d.com/


(R0m1 R) #4

Hi there !

I'm using agisoft photoscan in combination with Zbrush to clean my scans !
Most of them are done using a very cheap equipement from ikea smile

Some lamps to get a homogene lighting on your object
A turntable where you can stick a with paper with graduations


A nice white backgroud, I used this box with a big light behind to get a clean white backgroud.

then you put your object in the middle of your turntable and place you camera on a tripod.
In photogrammetry, the key is the sharpness of your photo ! So try to shoot at aperture f18/22 and iso 100 ! To have this conditions, a point and shoot or a smartphone camera won't be good. I will recommand a DSLR or an hybrid cam.

Then you take a photo each 20° angle and import the 18 photos into your photogrammetry software !
Most of the time I remove the object without changing the lighting conditions and the camera preset !
Why ? You will used it to do an automatic masking of your object on all your photos, it's a real time saver.
Most of them don't like turnatble since the backgroud is fixed, the reconstruction will failed so you need to mask the object, so that the soft can align photos only with informations on your object.

So finaly to sum up an old rig I've done:

to enlarge the image, click on it wink

As you can see, I used Zbrush to retopo the model and bake the normalmap with Xnormal.
Now you have a lighter model that is easier for sketchfab to handle.

Hope it have help some of you wink

And sry for my english ^^


#5

Awesome stuff. Great tips on using a turntable and masking. Do you just make the background transparent on every photo?

I usually try to decimate my scans in MeshLab [Quadric Edge Collapse Decimation (with texture)] But remapping + normalmap sounds much better...I'll have to learn...


(Bart) #6

Wow great info guys, thanks! I'll contribute my bit here as well:

My typical non-expert workflow uses Autodesk Recap360. I used 123DCatch before but it doesn't seem to work well anymore and returns errors every time I try.

  • I try to shoot my photos outdoors as much as possible to get the best even and bright lighting.
  • I'm applying some common sense to determine which locations might require some more photographs (with overhangs, for example).
  • After processing, I import my models in to Blender and use the Bisect tool to crop it back to a nice square model. I've tried several options here, but Bisect creates the cleanest edges. Here's a quick video on how to use that:

(Bart) #7

This would be a nice topic for a tutorial! Hmmmm :smile:


(Mauricesvay) #8

I usually use my Canon 6D DSLR with a 17-40mm lens. In a near future, I will probably try a narrower lens (50mm?) with a polarizing filter to remove some reflections. I usually shoot in RAW to keep details in shadows and highlights.
I then "recompress" the dynamic range at the center of the histogram before exporting to jpeg.
I finally use 123D catch or Recap360 to do the reconstruction.

I'm curious to see if adding markers on the subject (like stickers with b&w patterns) could help could improve the reconstruction success rate. Markers could easily be removed from the final texture I guess. Has anyone tried this?


(Bart) #9

Interesting - is there a way to get HDRI images from such a camera?


(Mauricesvay) #10

I guess you could, but you would still have to shoot multiple exposures.


(Andy lewis) #11

Tutorials!! Brilliant idea! I'll have to keep an eye out for them!


(Jeancharlesvarlet) #12

Hey @R0m1R, thank's for the tips wink About cleaning up using Zbrush, do you have an advice regarding Face scans ? What are the best things to use in Zbrush to make a face scan look better, especially when you want to print it ?


(R0m1 R) #13

Most of the time,my workflow is:
1- Import the mesh into Zbrush as a subtool
2-Duplicate your subtool (to keep the original)
3-Decimate the mesh to 500k using decimation master
4-Remove bad parts using masks or polygroups
5-Smooth the mesh using smmoth/smoothvalley/smoothpeaks brushes
6-Dynamesh the dubtool to remove bad polygons
7-Normally for 3D printing it will be sufficient


(Bart) #14

A friend of mine recommended using a LED light bar for indoor scans. In fact, he just taped it to his iPad for use with his Structure scanner wink He got himself a model that can adjust the color temperature for the best results.

Does anyone here have experience with those?


(Jeancharlesvarlet) #15

I will receive this baby in 2 days :

I'll keep you posted wink


Law and 3D Scans
(Gerpho) #16

Here's my contribution.
I use Smart3Dcapture (expert version). from Acute3D

I usually shoot with my Nikon D800, and focal lengths ranging from 14mm to 200mm. All work fine, even when mixed in a single project !
But I've also used the D300 (Beynac), and even my smartphone, although the picture quality of the phone isn't that good, and it appears in the scan : this sphinx was shot with my smartphone :

Sphinx statue - Hotel de Sully - Paris by Gerpho on Sketchfab


The key to success is to make sure the overlap is sufficient between shots. The specification is 15° angle or less between successive shoots, and I checked that greater values just don't work.
Appart from that, any sequence of pictures is OK. For instance : the roman arena was made by wandering through and around and climbing on the monument (plus a few photos shot from a crane). 300 photos later, here's the result (shot with the D800 and the 14-24 lens at various focal lengths ) :

Perigueux : remains of the roman arena by Gerpho on Sketchfab

This software can handle huge amounts of photos (depending on the version). I've scanned square kilometers at a few cm resolution, with good results (although far to big to be presented in sketchfab).

Everything's automatic, and you get a textured model with fantastic levels of detail, which can be exported to sketchfab in textured obj format.

Once it's finished, there may be a bit of retouching to do.
For the textures, I use Blacksmith3D (previously called "reality paint"), which is photoshop-like but on 3D, and very convenient to correct textures.
For the mesh itself, Autodesk's meshmixer is very easy to use, and doesn't mess up the textures (with the proper settings).
Both of these softwares readily import and export obj.
To be honnest the most difficult part when using all these softwares, is to discover, and get used to, navigation : some use LMB and different keys, some the mouse-wheel, some fonction-keys and when you switch from one to the other you end up trying to remember how to navigate. frowning

Both mesh and textures often need to be reduced to be proprely handled by sketchfab. To decimate, I use either meshmixer or blender. To reduce textures, photoshop is fine.
Finally, a zip file with the obj, mtl and jpg file, and here it is on sketchfab ! smile


(Bart) #17

Hey all,

Rowland Kirishima (AVATTA on Sketchfab) just shared a lot of information on his scanning setup in an interview on our blog. Enjoy!


(Humanti) #18

Hey there. I have a question for you. Blender can not find the normal map instructed where to look?

! K A L I B R A T I O N by humanti on Sketchfab


Tibor

(Bart) #19

Hi, I'm not entirely sure what you're asking. Can you elaborate?


(Bdnewnan) #20

For the most part I use Recap 360 by Autodesk. This is the upgrade from 123D catch also from Autodesk. I take a standard set of 360 degree photos, every 10 degrees at eye level, then take at least two more sets at a higher and lower angle. I try not to shoot into the sky as the lighting is bad. You have to capture all the elements of the figures, even the hidden areas. This is particularly true for complex figures. Its important to keep the camera on fixed settings, not automatic, and no flash. The camera should be mounted on a tripod if possible. When shooting dark figures, such as bronzes, then flat light is important, no shadows in the morning is best.
See my portfolio below
https://bdnewnan.sketchfab.me/folders/charles-umlauf