OSMGraphCLIP
Global location encoders learned from OpenStreetMap graphs (ACM SIGSPATIAL '26). Code: github.com/d-michail/osmgraphclip
Updated • 74 • 1Note Training data for the MS variants: multi-scale OSM graphs paired with coordinates for ~200k globally diverse locations.
d-michail/OSMGraphCLIP-MS-L10
Feature Extraction • UpdatedNote Multi-scale (graph + context rings), spherical harmonics L=10. Location encoder only needed at inference; no OSM data at query time.
d-michail/OSMGraphCLIP-MS-L40
Feature Extraction • Updated • 2Note Multi-scale (graph + context rings), spherical harmonics L=40. Location encoder only needed at inference; no OSM data at query time.
d-michail/OSMGraphCLIP-A-L10
Feature Extraction • UpdatedNote Graph only (no multi-scale rings), spherical harmonics L=10.
d-michail/OSMGraphCLIP-A-L40
Feature Extraction • UpdatedNote Graph only (no multi-scale rings), spherical harmonics L=40.
OSMGraphCLIP: Learning Global Location Representations from OpenStreetMap Graphs
Paper • 2606.08046 • PublishedNote OSMGraphCLIP: Learning Global Location Representations from OpenStreetMap Graphs. ACM SIGSPATIAL '26 (doi.org/10.1145/3841645.3842950). Code: https://github.com/d-michail/osmgraphclip