How vladimirescu/efficientgeometrysatellitesynthesis works
This repository is a command-line inference entry point for geometry-controlled satellite-image synthesis, rather than the implementation of the synthesis model itself. The user supplies a prompt and either coordinates or an OSM image; the script downloads the Hugging Face repository, loads its pipeline, invokes generation, and saves the generated image and control tile. The README documents coordinate-based OSM tile retrieval by the pipeline. The pipeline implementation is hosted remotely and is not present in the supplied tree, so its internal stages and invocation details are not expanded. Note: the README uses --osm_map, while the script defines --osm_path.
Main entry points: infer_HF.py Inference runner
Components 9
Inference entry
- CLI arguments
infer_HF.py - Inference runner
infer_HF.py
Geometry inputs
- OpenStreetMap API
- Local OSM image
Synthesis
- Hugging Face Hub
- SatSynthPipeline
Saved outputs
- generated.png
- osm_map.png
Other components
- User
Connections 11
- User → CLI arguments: provides arguments
- CLI arguments → Inference runner: configures
- Inference runner → Hugging Face Hub: downloads code and weights
- Hugging Face Hub → Inference runner: provides pipeline files
- Inference runner → SatSynthPipeline: loads and invokes
- Inference runner → Local OSM image: opens optional image
- Inference runner → SatSynthPipeline: passes prompt and coordinates
- SatSynthPipeline → OpenStreetMap API: fetches coordinate tile
- SatSynthPipeline → Inference runner: returns images
- Inference runner → generated.png: saves generated image
- Inference runner → osm_map.png: saves control image