A new arXiv preprint, "Geo-vision: Native-Resolution Geometry and Sampled Topology for Vehicle-Part Damage Classification," introduces a fixed representation for classifying damage to vehicle parts. The method integrates appearance with geometric and topological information, aiming to capture both visual and structural characteristics of damage.
The representation combines multiscale geometry, orientation, and sampled digital topology. It relies on topological descriptors such as connected components, holes, and Euler characteristics to encode the spatial structure of damage patterns. These features are intended to complement appearance-based cues.
The paper is available on arXiv under identifier 2610.23967. As a preprint, it has not yet undergone peer review, but it offers a concrete approach to leveraging geometry and topology for a practical classification task.