Our neural material [Rainer et al. 2019] Ours Reference Ours Multi-Res.Reference Multi-Res. [Rainer et al. 2020]
Log of Filter Kernel Size
Neural Offset
Neural texture lookupOur neural material Fig. 1. Top left: Our multi-resolution neural material representing Twisted Wool, rendered seamlessly among standard materials using Monte Carlo path tracing. The neural representation is trained using hundreds of reflectance queries per texel, across multiple resolutions, and is independent of the underlying input, which could be based on displaced geometry (in this example), fiber geometry, measured data, or others. Top right: The stages of our pipeline: computing a kernel size based on pixel coverage, evaluating a neural offset module for improved handling of parallax effects, evaluating a neural texture pyramid to obtain a local feature vector, and applying a small fully-connected neural network to obtain a reflectance value usable in a standard renderer. Bottom left: Comparison of our result to previous techniques and to a reference path-traced from the ground-truth geometry. Bottom right: Our results match the reference across resolutions. One additional lighting / camera angle shown.We propose NeuMIP, a neural method for representing and rendering a variety of material appearances at different scales. Classical prefiltering (mipmapping) methods work well on simple material properties such as diffuse color, but fail to generalize to normals, self-shadowing, fibers or more complex microstructures and reflectances. In this work, we generalize traditional mipmap pyramids to pyramids of neural textures, combined with