Sensitivity outputs improve inspection of differentiable renderers
Sensitivity as an Arbitrary Output Variable for Differentiable Rendering
Graphics
Summary
Differentiable renderers compute how changes in a 3D scene affect an objective, like an image feature, but their derivative outputs are hard to visualize and understand. The authors introduce sensitivity outputs that capture how much each scene parameter influences the objective. This allows easier inspection by creating views where users can see the effect of changes from different angles and levels of detail without recalculating derivatives repeatedly. Their approach treats these sensitivity outputs as regular render results that can be explored flexibly, similar to how images and layers are handled in graphics.
What this means in practice
- •For computer graphics engineers: Visualize and debug how scene parameters affect render objectives interactively from multiple viewpoints.
- •For visual effects artists: Inspect sensitivity maps over textured surfaces to refine and optimize scene details influencing render outputs.
Authors
Linas Beresna, Eugene Fiume
Abstract
Differentiable renderers expose the derivative of any scalar objective with respect to every scene parameter, yet unlike the primal image, which decades of arbitrary output variables (AOVs) have taught us to decompose, inspect, and composite, these derivatives have no established representation for human inspection. We introduce the sensitivity AOV, a render output carrying the sensitivity of an objective to the scene parameters that influence it. A single reverse-mode pass populates a sensitivity buffer over the scene's parameter hierarchy, from which many views are read rather than re-differentiated, in direct analogy to deferred shading: image-space sensitivity at object and parameter-type granularity, projections onto a freely navigable scene from any inspection viewpoint, and, for spatially varying parameters, per-texel fields carried to the surface through texture coordinates. We separate the fixed camera that defines the objective from the free camera used to inspect the result, and position reverse-mode attribution against its forward-mode dual. Our aim is not a single algorithm but a scaffolding that establishes derivative outputs as first-class render products alongside the primal image.