Lunar-G2R: Geometry-to-Reflectance Learning for High-Fidelity Lunar BRDF Estimation

TL;DR

Lunar-G2R tackles the problem of creating high-fidelity, spatially varying lunar appearance by learning per-pixel BRDF parameters directly from terrain geometry. The method uses a U-Net to map high-resolution lunar DEM crops to BRDF parameter maps and trains the network end-to-end with differentiable rendering (SurRender) against real orbital imagery, enabling SVBRDF estimation without multi-view or controlled lighting. On a geographically held-out Tycho crater region, Lunar-G2R reduces photometric error by about 38% compared with a Hapke baseline and improves PSNR, SSIM, and perceptual similarity, while capturing fine-scale reflectance variations. The work introduces a geometry-to-reflectance learning paradigm, demonstrates generalization to unseen viewpoints, and provides a large paired dataset to support future planetary reflectance research, with potential extension to other bodies and terrains.

Abstract

We address the problem of estimating realistic, spatially varying reflectance for complex planetary surfaces such as the lunar regolith, which is critical for high-fidelity rendering and vision-based navigation. Existing lunar rendering pipelines rely on simplified or spatially uniform BRDF models whose parameters are difficult to estimate and fail to capture local reflectance variations, limiting photometric realism. We propose Lunar-G2R, a geometry-to-reflectance learning framework that predicts spatially varying BRDF parameters directly from a lunar digital elevation model (DEM), without requiring multi-view imagery, controlled illumination, or dedicated reflectance-capture hardware at inference time. The method leverages a U-Net trained with differentiable rendering to minimize photometric discrepancies between real orbital images and physically based renderings under known viewing and illumination geometry. Experiments on a geographically held-out region of the Tycho crater show that our approach reduces photometric error by 38 % compared to a state-of-the-art baseline, while achieving higher PSNR and SSIM and improved perceptual similarity, capturing fine-scale reflectance variations absent from spatially uniform models. To our knowledge, this is the first method to infer a spatially varying reflectance model directly from terrain geometry.

Figures (13)

• Figure 1: Lunar-G2R estimates a spatially-varying parameterized BRDF of the lunar surface given the geometry. This estimated appearance allows for a realistic rendering of the moon surface (a), which closely matches the ground-truth observation (b).
• Figure 1: Dataset design pipeline
• Figure 2: Lunar-G2R inference pipeline for SVBRDF estimation from DEMs.
• Figure 2: Rusinkiewicz reparameterization of BRDFs. $\mathbf{n}$ denotes the surface normal, $\mathbf{t}$ the surface tangent, $\mathbf{h}$ the half-vector, and $\mathbf{w}_i$ and $\mathbf{w}_o$ the incident and outgoing directions. Adapted from Rusinkiewicz Rusinkiewicz1998.
• Figure 3: Examples of DEM--image pairs from the dataset: cropped DEM patches (left) and corresponding orthorectified lunar images (right).