Skip to content

Add MVTec AD preparation and baseline model decision - #3

Merged
Ray3R3 merged 1 commit into
mainfrom
feat/mvtec-prep
Nov 16, 2026
Merged

Ray3R3 merged 1 commit into
mainfrom
feat/mvtec-prep

Conversation

@Ray3R3

@Ray3R3 Ray3R3 commented Nov 16, 2026

Copy link
Copy Markdown
Contributor

Summary

  • scripts/prepare_mvtec.py: MVTec AD has to be downloaded after accepting its licence, so the script works from a local archive. It extracts only the requested categories, rejects unsafe archive paths and checks every defective test image has a ground-truth mask.
  • data/datasets/mvtec_ad.yaml: dataset manifest (source, licence, layout, categories). Data itself stays out of the repo.
  • docs/decisions/0001-baseline-model.md: PatchCore via anomalib as the baseline. Trains on defect-free images only, gives pixel-level score maps, and keeps the codebase Apache-2.0.

Testing

pytest (13 passed), including a synthetic archive with a missing mask and a path-traversal entry.

MVTec AD must be downloaded manually after accepting its licence, so
prepare_mvtec.py works from a local archive: it extracts only the requested
categories, rejects unsafe archive paths and checks that every defective
test image has a ground-truth mask.

ADR 0001 picks PatchCore via anomalib as the baseline: it trains on
defect-free images only, gives pixel-level score maps for localisation,
and keeps the codebase Apache-2.0.

Co-authored-by: Sam Oteng <98087809+samoteng@users.noreply.github.com>
@Ray3R3
Ray3R3 merged commit 07d7fd7 into main Nov 16, 2026
2 checks passed
@Ray3R3
Ray3R3 deleted the feat/mvtec-prep branch November 16, 2026 18:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant