Application of Disentanglement to Map Registration Problem
This addresses the challenge of coherent analysis for geospatial data users by enabling style-invariant registration, though it appears incremental as it builds on existing disentanglement methods.
The paper tackles the problem of aligning geospatial data from diverse sources by proposing a two-step image registration process that extracts invariant geographic content and matches based on it, using a β-VAE-like architecture with adversarial training to disentangle geographic information from visual styles and generate new map tiles.
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 $β$-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.