PROXY FACES

Masks, Manipulation, and Machine Vision

Proxy Faces is a provocative exploration of the fragility of modern facial recognition systems, even as machine learning grows increasingly sophisticated. Leveraging YOLOv7—a state-of-the-art object detection model—this project challenges the reliability of AI by designing physical "adversarial tools" that trick algorithms into misidentifying human faces. Starting with simple hacks like holding an ice cream cone or covering my face with translucent plastic, I discovered how minor alterations could confuse the model. But the real breakthrough came when I designed a wearable, rectangular mask devoid of facial depth, combined with magazine-cutout eyes and lips. Attached to modified sunglasses for practicality, this prototype successfully deceived YOLOv7, exposing how even advanced systems struggle with flattened, artificial features.