DS2 - ANM 694: WEEK 2
Assignment 2.1: Box Model the Head
Here's my progress for the week. Tried to stick to just the level / resolution of the demo videos for now.
Also included some step snapshots for fun.
INSTRUCTOR FEEDBACK
Very nicely done. Appreciate the step snapshots. Also as you add resolution work on defining details and definition. I'm fine with where it's at the moment. Think spacing of the edge loops looks even. Mouth can use more definition and details. Understand this is a work-in-progress. Like how this is progressing.
Assignment 2.2
Do some research on modeling with Polygons, NURBS and Sub Division surfaces. Share your findings on what the advantages and disadvantages are between them and which one you prefer and why.
POLYGONS, SUBDIVISION SURFACES & NURBS
If one works in games, polygons are king and probably always will be. Polygons are just triangles, by definition planar geometric entities that can be textured, lit and rendered quickly. Video cards are optimized to push polygons, trillions of them a second in some cases. The challenge with polygons is in representing organic shapes -- polygons are very well-suited for boxes but for spheres, not so much! How does one make a round object using only flat objects? The solution is to throw more polygons at it until they are so small you can't see that they're flat! In essence, that is what every software renderer does during the 'render' process, even if the source geometry isn't polygonal at all. From a workflow point of view, our polygon manipulation tools are just plain fast now, and with zbrush and subdivision modeling techniques we can create and control millions of polygons efficiently and effectively in real time. There are some areas where polygons, no matter how many you employ, simply do not suffice.
Nurbs (Non-uniform rational b-splines) are complex mathematical structures under the hood. They are by definition resolution-independent. Because the surface information described is so exact, they are required for models that require precise manufacturing. If one were to create a widget in polygons, no matter how many subdivisions one adds to that model, at some point the resolution will break down to faceting. Polygons only approximate a surface using triangles after all; nurbs use surfaces derived from bezier splines to describe surfaces. A comparable 2D analogy would be Illustrator versus Photoshop, or paths versus pixels. The biggest issue with nurbs is that they are very difficult and tedious to work with. They lend themselves to more mechanical applications, but complex, truly organic structures are nearly impossible to create, much less control, in nurbs. On the flip side though, they can be used as part of a polygon workflow to create clean, curve-based structures quickly for conversion to polys.
In terms of subdivision surfaces, there are a number of different implementations that should be considered. Back in the day, Lightwave was revolutionary for its metaNurbs modeling feature, (no relation to actual Nurbs and they later changed the name,) whereby one could work with a low-resolution cage that subdivided (and smoothed) the mesh on the fly. The mesh had to be all quad-based for this to work, but one could get to a very high-resolution and very polished end result quickly via this method. Having control over large numbers of vertices via one or a few vertices is a very powerful workflow we now have in spades via zbrush. In basic terms, subdivision surfaces is just subdividing each polygon (one becomes 4, iterating again each level) while averaging the positions of vertices according to a smoothing algorithm, kind of like polySmooth in maya. Keeping the operation 'live' is the key so that the source mesh can be continually edited with the result updating in real time.
Maya's Subdivision Surfaces implementation as introduced in the middle 2000's was a little different. In this implementation you can affect how the smoothing behaves by setting 'crease' values on the subdivision cage components. A creased edge will appear sharper with no additional vertices added to the control cage:
When converted to polygons you see that the smoothing algorithm is adding extra edge loops to 'hold' those edges crisp. Under the hood the shape created is as smooth as the renderer will allow. Interestingly, Pixar has implemented a subdivision component into Renderman whereby models are automatically subdivided as needed according to camera proximity. Models far away stay as lowest level to render quickly and ones close up subdivide as needed to always appear smooth.
NURBS AND CHARACTER MODELING
Back when Alias|Wavefront was developing Maya in the late 90's, later Academy Award winner Chris Landreth was working with them and using beta versions of the software to vet out features, workflows and techniques while also creating a short film. "Bingo" was this short film which you can see in entirety here:
Early versions of Maya were based on Alias Power Animator, which also didn't have much in the way of support for polygons. (Lightwave and 3D Studio / 3D Studio MAX gained a lot of market share by being polygon-based, cheaper and available on regular pc's). So Bingo was done entirely in nurbs. The challenge with nurbs for characters is that you simply cannot create a realistic head model without breaking up the surface into multiple pieces or "patches." These patches then have to join up in ways that when animated don't pull apart or show seams. Even when created perfectly, this can still happen.
Back to Bingo: In developing the characters for this short film, Landreth looked toward anatomy to construct his nurbs patches:
Landreth found that when the individual patches follow the natural anatomy and landmarks of the face, they deform in ways that help the patches and blend surfaces stay together.
The head above is from Bingo, (grabbed from The Art of Maya, 2007); amazing that the topology here so closely matches up to what we consider solid topology even today.
MODELING PREFERENCES
In terms of what types of modeling I prefer, it entirely depends!
For characters and most organic and simple things, basic box modeling either in maya or zbrush or more recently, back and forth! I never 'bonded' with maya's subdivision surfaces, but the ability to 'crease' and then export to zbrush with the creased edges preserved is pretty darn cool.
Like most of us I suffered through the Hard Surfaces class here at AAU. What a challenge! After weeks of tearing my hair out I did finally come to terms with nurbs and have really come to appreciate the fact that you absolutely cannot make a perfect, pinch-free curved surface in polygons no matter how much time and vertex tweaking you invest. For cars, mechanical objects or manufactured items that have to be perfect, nurbs is the bomb. Maya's nurbs implementation is pretty aged but it is generally functional; someday I'd like to learn the more full-featured Rhino with t-splines for designing jewelry and 3D printing applications. For everything else, polygons are just simply fast and super-flexible.
One of the biggest issues with nurbs is texturing; texture projections have to be more procedural and materials shader based. One does not have the type of control over texture placement for nurbs as polygonal UV's offer.
