Tech From Here: NL · Sources

A shirt that hides you from AI

Tech From Here: NL · Dutch Design Week 2026

The work this post covers and every source it draws on, each linked to the original.

Research cited

  1. Thys, Van Ranst, Goedemé arxiv.org
  2. Wu, Lim, Davis, Goldstein arxiv.org
  3. Xu et al., 'Adversarial T-shirt Evading Person Detectors in A Physical World' arxiv.org

Other sources

  1. simonweckert.com/digitalcamouflage.html simonweckert.com
  2. Digital camouflage (ddw.nl) ddw.nl

Images, slide by slide

Slide 1
Diagram: to people the wearer is plainly there; to a person detector in the artist's tests, no person is foundImage: RN Collins studio, code-drawn. Checked against Simon Weckert, Digital Camouflage, simonweckert.com (read 28 Oct 2026)
Slide 2
Diagram: three steps from camera image to a detector's score, and where a pattern pushes the score downImage: RN Collins studio, code-drawn. Checked against Thys et al., arXiv:1904.08653; Wu et al., arXiv:1910.14667 (abstracts)
Slide 3
Diagram: stop signs vary little, people vary a lot; the 2019 patch targeted peopleImage: RN Collins studio, code-drawn. Checked against Thys et al., arXiv:1904.08653 (2019), abstract
Slide 4
Bar chart: T-shirt attack success 74% digital and 57% physical, against 18% for the earlier physical methodImage: RN Collins studio, code-drawn. Checked against Xu et al., arXiv:1910.11099, abstract
Slide 5
Diagram: the limits the artist names, model, distance, angle and lighting, and what cannot be verifiedImage: RN Collins studio, code-drawn. Checked against Simon Weckert, Digital Camouflage, simonweckert.com (read 28 Oct 2026)

Follow RN Collins