Optimising for multiple shots : an analysis of solutions diversity in virtual camera composition

Abstract

Automatically generating cinematics in computer games could, not only increase the quality of the player experience, but also allow more cinematographic visualizations of the game for other audiences, such as game replays, machinima, audience views or game comics. In all these cases, the virtual camera controller is expected to find a high quality camera con figuration, often in as little time as possible; however, in several cases, one optimal camera configuration is not sufficient and multiple diverse cameras are needed. In this paper, we show how the intrinsic shape of the the camera composition objective function contains multiple alternative solutions, often visually different, that can be used to generate multiple different shots from the same description. It is already known that from an optimization standpoint, these problems are usually multimodal so that algorithms with restart and/or niching components are needed as we may otherwise miss the best optima. Here, we are especially interested in exploring the problem landscapes and see how the difficulty level of finding multiple good shots varies across several stereotypical scenes.