Field of flowers problem

We are cur­rently work­ing on a scene which will con­tains a large field of flow­ers and veg­e­ta­tion. Cur­rently our effects expert, Adib Souly is work­ing and test­ing dif­fer­ent sce­nar­ios in Sof­t­im­age XSI. You can see a sim­ple con­cept we did (this was done in a rush by Faraz Shan­yar) that was done to help us envi­sion the den­sity and dis­tri­b­u­tion of the veg­e­ta­tion we would need.

But the prob­lem was much thornier than we expected …

We knew from the begin­ning that there was no way in hell we could cre­ate and ren­der such a huge scene which con­tains cam­era ani­ma­tion, ani­mated shrub­bery and veg­e­ta­tion and mil­lions of poly­gons with­out some opti­miza­tion. We also had a model of a flower we had tested and needed to imple­ment in our scene in num­bers as you see in the con­cept. One other impor­tant fac­tor, which was very impor­tant to our tim­ing, was that the solu­tion had to be scal­able and portable since we had a total of at least 25 scenes which would need to imple­ment the solu­tion and the mere idea of doing it all by hand was painful.

Our first solu­tion was to use Sof­t­im­age XSI hair in dif­fer­ent lay­ers that would con­sti­tute our lev­els of detail. These LOD’s would be cre­ated by using a math­e­mat­i­cal expres­sion that would link the den­sity of our hair to their dis­tance from the cam­era. Unfor­tu­nately this did not give a very good result in our tests as the num­ber of hairs would not decrease or increase smoothly. Instead it cre­ated a sort of pro­ce­dural ani­ma­tion that we feared would be notice­able in our renders.

The next solu­tion was to use hair again but instead of ani­mat­ing the hair den­sity we decided to cre­ate a mul­ti­tude of hair patches which would have decreas­ing num­bers of hair strands and their posi­tion with respect to the cam­era would be con­trolled by weight maps which needed to be done man­u­ally for each scene. This could obvi­ously cre­ate some prob­lems with some of the scenes where the cam­era ani­ma­tion was more intense but we fig­ured that we could even­tu­ally used ani­mated weight maps if the need ever arose.

We decided to use light­weight objects as instances on the hair patches for the lay­ers that would be near the cam­era and sim­ple hair strands on lay­ers that are fur­ther out.

We still had prob­lems with ren­der­ing our scenes, we would run out of mem­ory or crash XSI or Men­tal Ray all the time so we knew we had to look for a sim­pler solution.

We even did a test on E-On Soft­ware Vue 6 but the frame took too long to ren­der and we didn’t have enough expe­ri­ence on Vue to be able to opti­mize it in time.

The next step in our opti­miza­tion was to opt for tex­tured sprites of ren­dered veg­e­ta­tion that would always face the cam­era still using the hair sys­tem. The advan­tage of such a method would be that we would use much less poly­gons and even lesser hair strands. In fact, we are still test­ing this stage and haven’t reached a final solu­tion but the incon­ve­niences of such a solu­tion would be the pos­si­ble low qual­ity of sprites near the cam­era or their 2D nature in extreme cam­era move­ments. There is a white paper done by Pixar on scene sim­pli­fi­ca­tion and opti­miza­tion called Sto­chas­tic Sim­pli­fi­ca­tion of Aggre­gate Detail which goes into great detail regard­ing this very prob­lem, but the truth is that this solu­tion was too tech­ni­cal for us to imple­ment and we’re not even sure that it could be eas­ily imple­mented in XSI due to it’s pro­ce­dural nature.

For the moment we haven’t reached a ver­dict yet, but I’ll post our results as soon as we have any.

---

related post

• Pipe Smoke Simulation
• Kaam game char­ac­ter finalized
• Samand auto­mo­bile render
• Super­hero Tutorial
• Blur Studio’s New Commercial

Line Break

Copyright 2010 - Vishka Studio. All Rights Reserved

• Projects
• 3D Models
• Concepts
• Shorts Stills
• About Us
• Contact Us