Coffee Mug, Fig. 17

The coffee cup and smoke are modeled with a single non-stationary GPIS. With sub-pixel covariance scales, spatial and ensemble averages produce similar macro-scale appearances. Ensemble light transport enables next-event estimation for improved importance sampling, as shown in the equal-time (20 min) comparison of unidirectional and MIS methods.

Our sparse convolution noise formulation further reveals fine surface and volumetric micro-structures through zoom-ins.

The viewer below uses JERI to compare high-dynamic range OpenEXR files directly in the browser. You can also use the LDR version using PNGs.