geoclip

Self-hosted companion for Immich that finds where photos without GPS were taken. Suggestions from photos taken close in time and from a local GeoCLIP model are shown on one map — nothing is written to Immich until you confirm.

Research toolkit for pixels-only image geolocation, retrieval, and evaluation

Self-hosted tool that predicts where any photo was taken, using the GeoCLIP neural network. Runs entirely on your own machine — no uploads, no telemetry. Writes ESTIMATED coordinates into EXIF.

A unified Python interface to Earth embeddings: image encoders, location encoders, and precomputed geospatial products.

Autonomous, keyless photographic geolocation using 5-pillar multimodal evidence fusion (OSM, 30m DEM skylines, satellite homography, building footprints). 100% local neural inference.

Multimodal image geolocalization via contrastive retrieval. Embeds ground photos, satellite imagery, text and GPS into a shared space (GeoCLIP-inspired), with cross-view retrieval and University-1652 transfer experiments.

Local, privacy-first photo geolocation: GeoCLIP + StreetCLIP + vision-LLM OCR clues + Mapillary visual matching, with a GPU-aware training pipeline