Application of Disentanglement to Map Registration Problem

TL;DR

This paper proposes a novel approach to map registration by disentangling geographic content from style using a combined $eta$-VAE and adversarial training, enabling style-invariant alignment of geospatial data.

Contribution

It introduces a new method that separates geographic information from artistic style in geospatial data using a combined $eta$-VAE and adversarial training approach.

Findings

Disentanglement of geographic content from style achieved.

Style-invariant map registration demonstrated.

Generation of map tiles with mixed styles possible.

Abstract

Geospatial data come from various sources, such as satellites, aircraft, and LiDAR. The variability of the source is not limited to the types of data acquisition techniques, as we have maps from different time periods. To incorporate these data for a coherent analysis, it is essential to first align different "styles" of geospatial data to its matching images that point to the same location on the surface of the Earth. In this paper, we approach the image registration as a two-step process of (1) extracting geospatial contents invariant to visual (and any other non-content-related) information, and (2) matching the data based on such (purely) geospatial contents. We hypothesize that a combination of $\beta$-VAE-like architecture [2] and adversarial training will achieve both the disentanglement of the geographic information and artistic styles and generation of new map tiles by composing the encoded geographic information with any artistic style.