A Review Focused Cloud Render System for 3D CGI Production
Authors
Karkov, Martin Fabrin ; Larsen, Kim
Term
4. term
Education
Publication year
2020
Submitted on
2020-06-12
Pages
66
Abstract
Det er udfordrende at kvalitetssikre 3D-CGI, fordi path tracing, en udbredt renderingsmetode, bygger på tilfældighed. Det betyder, at gentagne rendering kan variere og give uens resultater. Vi undersøger, hvordan man kan analysere path-tracede renderopgaver for at gøre gennemgang og feedback mere pålidelig. På baggrund af denne analyse designer vi et render-farm-system - et koordineret sæt maskiner, der renderer billeder (frames) - med funktioner målrettet selve review-fasen. Vi udvikler en minimal levedygtig prototype (MVP) af designet og evaluerer den i forhold til vores tidligere render-farm-system. Prototypen sænker omkostningerne og øger nøjagtigheden, samtidig med at den er let at styre og kan overføres til eksisterende render-farme. Resultaterne peger på, at systemet kan gøre den kreative reviewproces mere effektiv og tillidsvækkende.
Reviewing 3D computer-generated imagery (CGI) is hard because path tracing, a common rendering technique, relies on randomness, so repeated renders can vary and produce inconsistent results. We study how to analyze path-traced render tasks to make review more reliable. Based on this analysis, we design a render farm system - a coordinated group of machines that render frames - with features tailored to the review stage. We implement a minimal viable product (MVP) of this design and evaluate it against our earlier render farm system. The MVP reduces cost and improves accuracy, while remaining easy to control and compatible with existing render farms. These results suggest that the proposed system can make the creative review process more efficient and trustworthy.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Documents