Automated staging for virtual cinematography

Abstract

Camera 1, CU on George front screencenter and Marty 3/4 backright screenleft. George and Marty are in chair and near bar. (a) Scene specification (b) Automated placement of camera and characters (c) Resulting shot enforcing the scene specification Figure 1: Automated staging for a simple scene, from a high-level language specification (a) to the resulting shot (c). Our system places both actors and camera in the scene. Three constraints are displayed in (b): Close-up on George (green), George seen from the front (blue), and George screencenter, Marty screenleft (red). The white camera is the location of the computed camera.

An interactive staging-and-shooting solver for virtual cinematography

“…We test 3 configurations sharing most of their implementation to ensure a fair comparison: GPOs and backtracking strategy are identical. First, a baseline (B) uses [Louarn et al 2018] with minor modifications to allow dynamic scenes. The approach sees the problem as a cycle and performs a fixed-point computation with an ad-hoc order: area nodes first, then orientation.…”

“…A pruning operator O is represented by at least one directed arc A N s →N t O between a node N t and its dependency N s . This extends [Louarn et al 2018] to handle temporal relations.…”

“…Our approach extends [Louarn et al 2018] by generalizing the specification to include spatio-temporal relations, adding operators and a graph decomposition techniques to tackle the problem complexity, and adding several degrees of interaction on the computed results while maintaining a coherence vis-à-vis the specification.…”

“…Each entity X is parameterized by four constants: its width (w X , the radius of its bounding circle), its height (h X ) and its field-of-view angles: horizontal (Φ h X ) and vertical ( PLRS. [Louarn et al 2018] introduced Potential Location-Rotation Sets (PLRS), composed of a Potential Location Set (PLS) as a disjoint set of possible locations where an entity can be placed, and a Potential Rotation Set (PRS), a disjoint set of possible orientations the entity can have. Here, we build on these representations to encode the possible positions and angles where entities can be placed.…”

“…Techniques for placing and cutting between cameras have received significant attention in the computer graphics community [Christie et al 2008]. Yet, few contributions [Elson and Riedl 2007;Louarn et al 2018] addressed the problem of planning actor layout simultaneously with cameras and lights, which are all together necessary. Our work builds on a previous contribution [Louarn et al 2018] to extend specification and solver to handle spatio-temporal descriptions explicitly modeling and solving temporal relations between entities.…”