A Versatile Parameterization for Measured Material Manifolds

A popular approach for computing photorealistic images of virtual objects requires applying reflectance profiles measured from real surfaces, introducing several challenges: the memory needed to faithfully capture realistic material reflectance is large, the choice of materials is limited to the set of measurements, and image synthesis using the measured data is costly. Typically, this data is either compressed by projecting it onto a subset of its linear principal components or by applying non-linear methods. The former requires prohibitively large numbers of components to faithfully represent the input reflectance, whereas the latter necessitates costly algorithms to extrapolate reflectance data. We learn an underlying, low-dimensional non-linear reflectance manifold amenable to the rapid exploration and rendering of real-world materials. We show that interpolated materials can be expressed as linear combinations of the measured data, despite lying on an inherently non-linear manifold. This allows us to efficiently interpolate and extrapolate measured BRDFs, and to render directly from the manifold representation. To do so, we rely on a Gaussian process latent variable model of reflectance. We demonstrate the utility of our representation in the context of both high-performance and offline rendering with materials interpolated from real-world captured BRDFs [MPBM03b].