cff-version: 1.2.0
message: "If you use this software, please cite it as below."
title: "VEIL-PGD: Vision-Encoder Imperceptible Layer for Training-Data Protection"
abstract: >-
  A research preview of an image-side protection method that applies a
  low-visibility, EOT-hardened, ensemble momentum-PGD adversarial perturbation
  to public images so that vision and vision-language training-data ingestion
  pipelines derive semantically displaced captions, tags, and labels.
authors:
  - alias: cebeuq
version: "0.2.1"
date-released: "2026-07-08"
license: LicenseRef-PolyForm-Noncommercial-1.0.0
url: "https://huggingface.co/cebeuq/veil-pgd"
repository-code: "https://huggingface.co/cebeuq/veil-pgd"
keywords:
  - training-data-protection
  - adversarial-examples
  - vision-language-models
  - CLIP
  - anti-scraping
  - PGD