3/9/2008
The Procedural Temple

The first project is shaping up. Houdini's strength is its highly structured and procedural approach. People coming from the Maya / 3DS camp might not exactly feel comfortable with Houdini, but as computer scientist I feel immediately at home. So my first project is a greek temple model. But not just a model, but a generic parametrised model. Parameters include the various measurements (length, width, height, height of roof, offsets etc.) and construction detail (number of columns, etc.).
I could have implemented an automatic model generation tool using MEL in Maya, but I doubt I would managed that quickly, especially not if the parameters should be driven by animation curves later on...
So two evenings later, I have a Houdini Asset that generates a temple model and is fully parameterisable. I'm already in love with Houdini.

It's saturday afternoon, I have a bit of spare time. Time to add some life into the temple asset. Animating the asset turns out to be trivial. Just set key frames for the parameters, that's it. Defining the animation took less than an hour, and rendering it out was surprisingly quick. Admittedly, the animation won't win any prizes for artistic content, but it demonstrates the point.

The exercise brought some interesting revelations:
1) A parameters defined as "int" behaves exactly like a floating point number during animation. Rounding code was therefore necessary to make my columns "pop" into position instead of smoothly shifting around... An int is a float - erm, yes, I have to get used to that...
2) Shadow maps rendered with Houdini Apprentice also include the watermark... Erm, no, that's not a good idea - Ray traced shadows it is, then...

The temple needs some life inside. So the next task is to establish the link between Poser 7 and Houdini. Poser exports animations as a sequence of geometry files, one for each frame. The format understood by both apps is OBJ. Collada would have been nicer, but Poser 7 doesn't export animation data there, a limitation of the beta exporter I used (time to check out the latest updates, I guess...).

Several issues surface here:
1) Houdini does not import any material definitions from OBJ files. This is pain number one.
2) Poser does not export objects for every body part, even if they use different materials. This pain number two.

Material definitions are only imported for collada files, but I can't use that path. In any case, textures are ignored - although they'd be simple enough to fix later. Instead I have to assign materials manually to groups. Issue 2) further complicates things: I have to split the objects into the individual material components. The eyes, for example, consist of the eyeball and the iris, yet they are exported as one group only. Similarly, eyelashes are part of the face group. It took a bit of struggling to find a reasonably elegant solution to identify the parts, but I eventually managed.

Above I mentioned animations. Houdini made importing the sequences of OBJ files extremely simple: There's nothing needed to do - it automatically loads the next geometry for the next frame. Thanks to the procedural approach to applying materials and identifying the groups, I only have to define the workflow for a single instance of the geometry, materials are the assigned to the following frames automatically. I love this.

The exercise also revealed some issues in Poser. The automated walk tool is cool, but the results are not beyond some doubt. Luckily, my goal is to learn about the process, not generate the next great animation...

Shading network

Walk animation

Until now, I used just the very basic materials for the figure: Phong shading plus texture. But that doesn't get one very far. Eyelashes, for example, have a separate alpha file. So, which shader to choose (short of writing my own). I tried the "VEX Super material" ubershader, but for some reason the texture coordinates for alpha don't work with the obj model. So I settled on the FBX shader that approximates the Maya default shader network to a good extent. All in all, it took a bit of fiddling around, but the shader works so far - until I enable the bump map, which overall works nicely, but has a few singularities where the normals go crazy. Ugh.

Walking V1

The Poser animation clearly needs improvement, and I also managed to mess up the shader parameter interface. Still work in progress, but the walking path now looks a bit better after tweaking the Poser interface, and I seem to have fixed the shader for the moment. Changing the animation in poser also changed the order of primitives in the obj file, so the identification code for the elements has become slightly more sophisticated. I missed the indicator for balance and target, so it looks as if there were some mice running around. I can fix that another time...

The overall atmosphere has also profited a lot from optimizing the lighting - not finished yet, it's still too much of an incandescent lighting style instead of torches, and too bright in general. There isn't much of a story there - but hey, I'm a software engineer, not an artist... ;-)

Over the last few days, I've populated the temple with additional props. Just adding a few props to the temple make a big difference, it suddenly looks a lot less artificial. Lighting has been optimised as well, with intensities being modulated by a noise function to make it all look a bit more like torches.

Rendering out the animation has become a bit of a pain. Although I optimised the materials (no reflections - ray-tracing seems awfully slow on Houdini, and the visual difference on the objects was minimal), it still takes quite some time to render, but the bigger issue now is stability, Houdini crashes after a few hundred frames on this scene. Oh well...

That concludes the Temple series for the time being.