# SIFT mechanism reference

Run `python examples/sift.py` for deterministic self-checks or `python examples/sift.py --json` for parity data. Python 3.9+; standard library only.

## Source and provenance

- Authors: David G. Lowe.
- Version: International Journal of Computer Vision 60(2), 91–110 (2004), author PDF.
- Primary source: https://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf
- PDF: https://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf
- PDF SHA-256: `06f8acc68583e8b7a041d1fda84db83283e49f6cedc4638ea5cdd2f61caf8e3e`
- Source crops: `public/images/papers/sift/source-crops.json` records one-based PDF pages, PDF-point rectangles, render scale, dimensions and image hashes. These are attributed cropped/rasterized excerpts; no open license was verified.
- Reference authored for Abhik Sarkar's review, 2026-11-06. Outputs are calculated teaching values, not recorded checkpoint predictions.

## Scope and limitations

Models strict scale-space neighborhood extrema, Hessian curvature-ratio rejection, L2/clip/L2 descriptor postprocessing and unsquared-distance ratio matching. Input neighborhoods and histogram arrays are authored. Does not construct Gaussian pyramids, refine extrema, compute full Gaussian-weighted trilinearly interpolated descriptors, match actual images, estimate poses or reproduce published benchmarks. The strict ratio boundary follows common implementation convention; the paper prose rejects values greater than 0.8.

## Reproducibility

`node --test test/sift.test.mjs` checks independent mathematical invariants and JavaScript/Python parity. Canonical source lives in `examples/sift.py`; named region excerpts and public downloads are regenerated by `scripts/sync-paper-examples.mjs`. Do not edit generated mirrors.

To verify crop bytes, obtain the pinned PDF and run `uv run --with pymupdf python scripts/render-paper-crops.py --manifest public/images/papers/sift/source-crops.json --pdf /path/to/paper.pdf`. PyMuPDF is an authoring tool, not an application or reference dependency.
