This water bottle looks like a water bottle to humans, but has been designed to minimize the probability of being correctly identified by computer vision algorithms.* In the post-apocalyptic future, we'll only be able to survive the robot uprising by hiding our water -- something we all need to survive. You can also hide your hot chocolate in it, if that's what you need to survive.
Adversarial patches don't always transfer well between models. This patch was specifically designed to be effective against YOLOv2.
• High-grade stainless steel
• 17 oz (500 ml)
• Dimensions: 10.5″ × 2.85″ (27 × 7 cm)
• Vacuum flask
• Double-wall construction
• Bowling pin shape
• Odorless and leak-proof cap
• Insulated for hot and cold liquids (keeps the liquid hot or cold for 6 h)
• Patented ORCA coating for vibrant colors
• Hand-wash only (dishwasher not recommended due to vacuum seal)
* This means that it might get classified as a vase or a plant, depending on how far away the camera is from the water bottle, and what the viewing angle is.
This adversarial designs water bottle uses an untargeted adversarial patch that minimizes the class probability for bottle in the bounding box for the object. The patch is trained with dynamic spline warping to account for perspective shifts and the curvature of the bottle's surface. You can read the original adversarial patch paper at https://arxiv.org/abs/1712.09665, and learn about the warping method pioneered on adversarial t-shirts at https://arxiv.org/abs/1910.11099.