Ray Gardener wrote:
> Hi all, thought I'd share the output of Leveller's recent changes to its 
> POV-Ray export -- textured billboards that auto-rotate to face the 
> camera during stills and animations, plus path-based camera motion.

One thing you could try is to render a scene with just the 3d model of 
the tree from the same general direction as the camera will view the 
actual tree in the final scene; then use this image as the visible 
billboard representation of each tree in the scene.  That will tend to 
mitigate the "Doom corpse" effect, IYKWIM.  For each frame, render the 
tree from that frame's view direction.

The model will be the only thing in the scene and should render quickly.

Also, the shady side of the tree will be darker than the sunny side of 
the tree.

For the shadow, you would use one consistent image_map for all of the 
frames, rendered with a camera where the sun would be, against the 
ground as background.

The upshot is that each frame is done in a two-render pass, one for the 
image_map for the visible tree objects, and then the final render for 
the frame.

For additional fun and games, you could also make specially-blurred 
image_maps from the tree renders, so that when they are used in the 
final render they are automatically anti-aliased and will not flicker in 
the animation.  I used this in my IRTC entry "Robotany" to make some 
lettering smoothly legible without insane levels of anti-aliasing.

Regards,
John

Post a reply to this message

Alain wrote:
> Now, the only thing that can tip you is that the bilboards don't self 
> shadow. Most peoples won't notice that, it's minimised by the actual 
> texture and the shadows from the other ones.

For trees, yes, it's not too bad. I'm now working with grass and, 
depending on the variation of blade color and height, the lack of 
self-shadowing in grass is much more noticeable.

Overall, I stress to end users that these are for prototype rendering. 
Given that the "age of multicore" is getting nicely underway, I don't 
expect to compete directly with proper rendering but rather to augment 
it. Especially for animations, e.g., in high-end studios you'll see them 
render out fast or realtime OpenGL clips, which look bad, but are great 
at conveying the location, shape, motion, etc. of elements. JvS's 
two-pass suggestion for billboard generating is interesting in that it's 
a fast and easy way to preview CSG objects -- all the complexity is 
flattened into an imagemap.

Granted, this competes with simply using low-quality settings, but 
having the option could be useful, because the type of quality tradeoff 
is quite different.

Ray

Post a reply to this message

John VanSickle wrote:
> Ray Gardener wrote:
>> Hi all, thought I'd share the output of Leveller's recent changes to 
>> its POV-Ray export -- textured billboards that auto-rotate to face the 
>> camera during stills and animations, plus path-based camera motion.
> 
> One thing you could try is to render a scene with just the 3d model of 
> the tree from the same general direction as the camera will view the 
> actual tree in the final scene; then use this image as the visible 
> billboard representation of each tree in the scene.  That will tend to 
> mitigate the "Doom corpse" effect, IYKWIM.  For each frame, render the 
> tree from that frame's view direction.
> 
> The model will be the only thing in the scene and should render quickly.
> 
> Also, the shady side of the tree will be darker than the sunny side of 
> the tree.

Yeah, I'd like to try that. My tree models aren't too complex, should 
only add a few seconds per frame. I can also skip imagemap creation 
every N-1 out of N frames given some angular tolerance, say, wait until 
the camera angle has gone above five degrees from the last time we 
rendered imagemaps.

Ray

Post a reply to this message