1/27/2005
   slide 13
Digital Michelangelo
Project
•Create an explorable 3D model
•Stanford and University of Washington
•Custom Cyberware statue scanner
•
The Digital Michelangelo Project
http://graphics.stanford.edu/data/mich/
 
So, not too many computer-related issues in the last example, but… now let’s look at impact of reproduction at a whole new level…

 Michelangelo, when he was painting the Sistine chapel really wanted to be creating sculpture—he was forced to paint it against his wishes.
One of his most famous sculptures is his David.
Goal of Digital Michelangelo Project was to make a 3D model that people can explore from any point of view, different lighting, etc.

Visual Culture:
History: It’s David from David Goliath story in the bible.
The statue of David was started by a different artist, Agostino di Duccio, in 1463. He picked out a rather narrow piece of stone, which was customary for artists of his denomination. If you are an art expert, you can see from the side that this is not a piece that Michelangelo would have picked. It is too thin. http://vlsi.colorado.edu/~rbloem/david.html
Politically: resonated with Florence and its situation within the city state of Italy
Also, aesthetically very beautiful sculpture.

Already several full-scale replicas in Florence.

Computer Graphics and Visualization (CS): The Stanford project, led by Marc Levoy, was a heroic proof-of-concept project costing millions of dollars.
Not replicable/scalable at this point. Going to tell you a bit about it because: fascinating project, demonstrates some of the many implications of visual computing for fields from art to tourism to philosophy, as well as technical issues in computer graphics (papers on project published at SIGGRAPH and elsewhere).
•3D scanning not new: People scan 3D stuff all the time, cheap devices, used for movie special effects, etc., etc. but really good quality, accuracy of large-scale delicate rare unmovable art works is a different story.

•Sampled sculpture at less than 1mm intervals (it’s over 5feet tall)

•Scanned with customize device but some parts of surface self shading—had to use hands canners. Still spent hours trying to fill in missing data—holes in the final model.

•Dealt with
--had to hire 24 security guard
--had to have shifts and scan all night to get done in time frame of grant and italian gov.
--had to use manual controls and encase parts of the scanner in foam to ensure no accidental damage to sculpture
--had to account for light falloff from actual lighting to create accurate model of surface that could be lit any way a user desires in a 3D scene
--data sets huge (many gigabytes) and hard to deal with.


•2mm accuracy
•4 million polygons—triangles-- in the head.
•8 million for entire statue.

•Raw data was over 2 billion triangles

•Head takes 12 hours to render.

•Veining and reflectance artificial. Renderings include physically correct subsurface scattering, but with arbitrary parameters.
•Data for flawless, watertight, full-resolution model of the entire statue, with color = 2 billion polygons and 7,000 color images, occupying 32 gigabytes.
•3D graphics in general incorporates most of the concepts learned in 2D graphics—the texture maps and “7000 color images” referred to are all raster data—something we’ll be learning about in a few weeks. Artists , scientists, art historians, and others will need to know the basic terminology and concepts in this field.

Harder to make this data useful than originally thought—needed cleaning” and datatsets are huge. http://graphics.stanford.edu/data/mich/
Digital Michelangelo Project
Model issues: noisy data, misaligned meshes, holes.
Sounds conceptually simple but very challenging—equipment, logistics, angle of laser when scanning, scattered light from marble surface, dirt on sculpture, limits on dataset size (huge but could have captured more info, but dataset would have been unmanageable). Vibrations,
<1mm to 5feet tall for david. Huge dynamic range 20,000 to 1.
“However, reconfiguring the pan-tilt [scanning] assembly proved more problematic. In retrospect, this should not have surprised us; 11 microns of play - 1/10 the diameter of a human hair - in a pin and socket joint located 5 cm from the pan axis will cause an error of 0.25 mm at our standoff distance of 112cm. In general, we greatly underestimated the difficulty of reconfiguring our scanner accurately under field conditions.
Range scanning. A typical range scan consisted of several concentric curved shells separated by translational motion of the scan. We often spent hours positioning the gantry in fruitless attempts to fill holes in our model of the David.
Issues of laser possibly damaging statue –no.
But did have to use manual controls for automatic scanner head movement to ensure no damage. scan head encased in foam rubber.