Determine whether shading supervision improves physical terrain resolution

Establish whether the differentiable Lunar–Lambert shading objective, used as an image-derived constraint in the MarsFM single-image Martian relief-estimation model, improves physical terrain resolution.

Background

MarsFM supplements stereo-derived relief and differential-geometry supervision with a differentiable photoclinometric loss that renders predicted relief using a Lunar–Lambert reflectance surrogate and compares the render with the co-registered HiRISE RED orthoimage. The objective is intended to exploit image-grid information beyond the coarser posting of stereo-derived terrain references.

The paper explicitly notes that the surrogate does not model spatially varying albedo, cast shadows, or the complete bidirectional reflectance function. Consequently, image agreement may reflect appearance effects rather than genuinely recovered terrain structure. The reported experiments do not isolate the shading term through a matched ablation or validate physical terrain resolution against independent high-resolution references, leaving its actual contribution unresolved.