Overview
- Simon Weckert unveiled a garment called “Digital Camouflage” that he developed using an iterative adversarial loop to confuse object detectors.
- Weckert refined the pattern by testing it against the open-source YOLO person detector and says the final design lowers the model’s confidence in recognizing a human.
- Independent media tests reported by Gizmodo reproduced the effect on YOLO, providing corroboration while leaving wider effectiveness unproven.
- The pattern works by altering the statistical visual cues AI models use to assemble body parts into a single person, not by hiding the wearer from human sight.
- The project was shown in Berlin’s Kottbusser Tor as a critique of planned AI video surveillance and joins a small market of anti-surveillance clothing while defenders could respond with model updates or regulation.