Written By Jyotika Prasad

ENG + TECH

written by JYOTIKA PRASAD

Five Minute Trees

Modeling 3D Content in Minutes

Have you tried to draw a tree recently? For many of us, such efforts usually result in a collection of ragged curves sprouting from two straight lines. Now, using innovative software developed by the Stanford Virtual Worlds Group, it is possible to design a realistic, threedimensional tree in less time than it will take you to read this article. Dryad is the initial result of an endeavor to improve computer modeling of real world objects, a process otherwise known as 3D content creation. “In traditional design techniques you have to model things from scratch,” says Dr. Vladlen Koltun, Assistant Professor of Computer Science at Stanford and head of the Stanford Virtual Worlds Group. With current tools, modeling a natural-looking tree would require “spending a day learning and another day modeling.” In contrast, Dryad employs “an intuitive approach to 3D modeling” that lets a novice user create a tree in less than five minutes.

Collaborative Mapping of Parametric Design Spaces

Dryad is a prototype for a new design technique called Collaborative Mapping of Parametric Design Spaces. “Parametric design space” refers to a class of real objects that can be approximated virtually using a single set of numbers. A tree can be specified by its height, trunk thickness, number of branches, distribution of leaves,

42 stanford scientific

and other similar parameters. Each point in this space corresponds to a particular choice of values for the parameters and represents a unique object. Dryad uses this parametric design space to model trees. It presents the user with a two-dimensional map of the tree space which can be navigated using an interface similar to that of Google Maps. The map represents a subset of the potential trees that can be modeled, with trees in the same “neighborhood” having similar values for their parameters. As for Dryad’s creative aspects, given a particular tree, the user can produce a similar yet distinct tree simply by tweaking its parameters. Alternatively, the empty space between two trees can be selected to interpolate between them and create a hybrid of the two. Combining a tall tree and a short tree would generate a tree of medium height, resulting in a traceable progression of parameters from one extreme to the other.

Challenges in Design and Development

While this may seem simple conceptually, developing the software has presented some serious challenges. Although they are presented as points on a two-dimensional map, trees in Dryad are actually specified using 98 parameters. “The geometry of your average tree is so detailed, and you are literally getting a tree every-time you

move a pixel. The geometry is completely changing. [It was hard] just dealing with the technical implications of hundreds of thousands of polygons constantly changing whenever you moved your mouse,” explains Daniel Gibson, a junior majoring in Computer Science who worked on Dryad. Even once the software was developed, “figuring out how to provide the user a way to navigate through the many-dimensional space was a really hard problem,” says Koltun. Although the aim of the software is to create trees, in fact very little of the tree space contains points that resemble anything close to a tree. A point picked at random from this space will more often appear like a spiky mutated blob. In order to provide the user with actual trees as a design basis, “collaborative mapping” allows the software to connect to a central server and thereby learn about every tree that has been created by other users. Dryad samples points from this space with a bias towards areas surrounding points already chosen by users, and thus presumed to contain trees, ensuring that the initial map presents reasonable looking trees. Finally, when the user creates and saves a new tree, it is uploaded to the server to aid future exploration of the space. Users all over the world collaborate in charting out a guidebook for the tree space. As more people use the software, it becomes more and more refined.

Credit: Stanford Virtual Worlds Group

“I think the most exciting part for me is how broadly applicable this is. The prototype we have just enables tree creation, but now what we are working on is opening it up to any design space at all. We are on the verge of creating a much more exciting and broadly applicable tool.” - Daniel Gibson

Since Dryad was initially launched in December 2007, over 13,000 people have downloaded the software. “[The user response] confirmed our notions to a degree that surprised us. Users created these beautiful trees that we didn’t know existed in our design space,” says Koltun. “My friend’s three year-old cousin was modeling trees,” says Jerry Talton, one of Koltun’s PhD students and a member of the Dryad team. Since parametric spaces are ubiquitous, the design principles behind Dryad can be applied to many real-world objects. As Gibson states, “I think the most exciting part for me is how broadly applicable this is. The prototype we have just enables tree creation, but now what we are working on is opening it up to any design space at all. We are on the verge of creating a much more exciting and broadly applicable tool.” “[The idea can be used for] anything that is drawn from a parametrizable domain, such as faces, buildings, other kinds of vegetation and automobiles,” reveals Koltun. “Maybe there’s a parametrizable model for body languages...to tell my avatar how I want it to move,” adds Talton.

Other objects could involve more parameters than trees, but Koltun believes that the technique will scale to more dimensions. “If anything had to break in our dimensionality reduction process, it would’ve broken already. One hundred dimensions is already beyond the ‘easy’ threshold.” Dryad and its future incarnations could help address a sore point for most users of virtual worlds and other graphics environments. 3D modeling still remains an esoteric domain, depending largely on trained professionals and complex software. Dryad challenges these constraints by facilitating rich 3D content creation by lay users, allowing a much wider and more democratized range of expression. Says Koltun, this “enables creativity of a sort that just doesn’t exist today”.

A small sample of the variety of designs achievable with Dryad.

Credit: Stanford Virtual Worlds Group

The Future of Dryad

JYOTIKA PRASAD is a masters’ student in Computer Science. Aside from writing, she also likes dance, theater and information retrieval.

To Learn More

For more information, visit the Dryad website at http://dryad.stanford.edu

volume VII 43