# VGG mechanism reference

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

## Source and provenance

- Authors: Karen Simonyan, Andrew Zisserman.
- Version: arXiv:1409.1556v6, 10 April 2015; ICLR 2015.
- Primary source: https://arxiv.org/abs/1409.1556v6
- PDF: https://arxiv.org/pdf/1409.1556v6
- PDF SHA-256: `83728f9efc21081792902b4c17a4657022d1e4a92ec81655c2427f4ef0755e50`
- Source crops: `public/images/papers/vgg/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

Calculates original D/E shape traces, theoretical receptive fields and exact parameter counts including biases for 1000-class classification. Kernel-stack comparisons hold channel width constant and exclude biases. No checkpoint, image predictions, training, latency measurement or benchmark reproduction is supplied. Original VGG has no batch normalization; VGG-BN and global-average classifiers are different variants.

## Reproducibility

`node --test test/vgg.test.mjs` checks independent mathematical invariants and JavaScript/Python parity. Canonical source lives in `examples/vgg.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/vgg/source-crops.json --pdf /path/to/paper.pdf`. PyMuPDF is an authoring tool, not an application or reference dependency.
