themisim

Similarity search over THEMIS all-sky imager (ASI) auroral imagery. A SimCLR-trained ResNet-18 encoder turns each 256×256 frame into a 512-D feature vector whose cosine proximity tracks visual and morphological similarity rather than pixel-level identity; those vectors are indexed with FAISS (OPQ-IVF-PQ). Given any indexed frame, themisim returns the most visually similar frames across the archive.

Contents

• Quickstart
  • Query an existing index
  • From the command line
  • Build an index from scratch
• API reference
  • Querying
  • Exporting results
  • Building an index
  • Visualization

Installation

pip install themisim

faiss-cpu is pulled in automatically (the query path is CPU-only). A CUDA-enabled PyTorch speeds up index building but is optional — every stage falls back to CPU.

Indices and tables

• Index

• Module Index

• Search Page