Constant memory method improves gradient estimation in light tracing
Stochastic Graph Compression for Constant-Memory Differentiable Light Tracing
Graphics
Summary
Tracing the paths that light takes to create images is very complex and usually uses a lot of memory, especially when calculating how small changes affect the image. The authors created a new method called ResLRB that keeps track of only one important part of each light path, reducing memory use while still giving accurate results. This method uses randomness cleverly to pick representative points and then reruns the path calculations when needed. This approach helps graphics programs predict subtle changes in lighting without running out of memory.
What this means in practice
- •For graphics software engineers: Enable memory-efficient gradient calculations in light tracing renderers to improve differentiable rendering workflows for scene optimization.
- •For augmented reality developers: Support real-time differentiable lighting effects on devices with limited memory by applying constant-memory gradient tracing techniques.
Authors
Linas Beresna, Eugene Fiume
Abstract
We describe Reservoir Light Replay Backpropagation (ResLRB), a method for reverse-mode differentiable light tracing whose memory is constant in path length and in the number of sensor connections per light path. Naive automatic differentiation of a light tracer records a computation graph that grows with the number of valid sensor connections along each path, since flux can be splatted from every scattering vertex; adjoint path replay removes the analogous dependence on depth for viewpoint path tracing, but its non-branching structure does not extend to the splatting case. We compress the adjoint graph during the primal pass by stochastically retaining a single representative sensor connection per light path via a streaming weighted reservoir. The adjoint pass reconstructs the path deterministically by replaying its pseudorandom number sequence and backpropagates only through the retained connection, yielding an unbiased gradient estimator whose peak memory is that of differentiating a single scattering event.