In 2025, artist Simon Weckert presented Digital Camouflage, a garment using an adversarial pattern to make AI person-detection systems fail to recognize its wearer. Ars Electronica describes the textile as designed to confuse computer-vision algorithms used in surveillance, security, and autonomous systems, while Hypebeast reports that TC-EGA generates its tileable pattern.
In 2025, artist Simon Weckert presented Digital Camouflage, a wearable textile project that applies adversarial-machine-learning techniques to AI-based person detection. Ars Electronica describes the garment as designed to make computer-vision systems fail to identify a wearer as a person.
The clothing appears as an abstract graphic print to human viewers. Its stated target is not the camera itself, but the object-recognition model processing the camera feed. Ars Electronica identifies relevant applications as surveillance, security, and autonomous systems.
A garment-scale adversarial example
Adversarial attacks manipulate an input so that a machine-learning model produces an incorrect result. In this case, the input is the visual appearance of a person wearing the textile. The project adapts that idea from digitally altered images and localized physical patches to a pattern covering the garment's surface.
Hypebeast reports that the design uses a continuous "Adversarial Texture" rather than a fixed printed patch. Its account identifies TC-EGA as the generative AI method used to produce a tileable textile pattern, and describes the intended function as generating false visual features across changes in camera perspective and fabric folds.
That whole-garment approach addresses a practical weakness commonly associated with physical adversarial examples. A localized patch can leave a detector's field of view when a person turns, moves, or partially occludes the pattern. Hypebeast characterizes earlier patch-based approaches as vulnerable when fabric folds or camera angles shift; Digital Camouflage instead distributes the pattern across the clothing surface.
Yanko Design reports that the collection is manufactured in Latvia from a textile blend of 65% recycled polyester and 35% polyester, with the pattern applied through digital textile printing.
Limits of the claim matter for deployment
The project is best understood as an adversarial-example demonstration against particular classes of AI person detectors, not as invisibility from cameras or a universal method for bypassing surveillance. A camera can still capture the wearer even if a downstream detector fails to produce a person classification.
In deployed computer-vision systems, the transferability of physical adversarial patterns can vary with detector architecture, model training data, resolution, compression, lighting, viewpoint, motion blur, and multi-camera tracking. Systems combining person detection with pose estimation, re-identification, face recognition, or human review can also create different failure conditions from a single object detector.
Security evaluations of comparable wearable attacks typically need to test at least three dimensions:
- •Detection performance across multiple models, versions, and confidence thresholds.
- •Robustness across illumination, distance, pose, occlusion, movement, and textile deformation.
- •Transferability from controlled tests to real camera pipelines, including compression and post-processing.
For ML practitioners building physical-world vision systems, the work illustrates why robustness testing cannot stop at clean benchmark imagery. Comparable adversarial examples expose a gap between image-level model accuracy and behavior in open-world sensing environments, where inputs can be deliberately designed to exploit a detector's learned visual features.
Key Points #
- 1Digital Camouflage applies an adversarial texture across a whole garment, targeting AI person detection rather than camera hardware itself.
- 2Hypebeast reports TC-EGA creates tileable patterns intended to remain disruptive across fabric folds and varying camera viewpoints.
- 3In computer-vision deployments, wearable adversarial examples require testing across models, lighting, motion, compression, and multi-stage analytics pipelines.
Scoring Rationale #
The project is a relevant physical-world adversarial-ML example for teams building person-detection and surveillance systems. Its practical impact is constrained because the retrieved materials do not report detector benchmarks, target-model details, or independent robustness testing.
Sources #
Public references used for this report. Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
[Active Search Campaigns by BudgetEasy](/problems/sql/active-search-campaigns-by-budget)
[High CPC Clicks & Poor Landing PagesMedium](/problems/sql/high-cpc-clicks-poor-landing-page)
[Campaign ROAS by Attribution ModelHard](/problems/sql/campaign-roas-by-attribution-model)
250 free problems · No credit card