Unadversarial Examples: Designing Objects for Robust Vision

TL;DR

This paper introduces a framework for designing objects that are explicitly optimized to be confidently detected or classified, thereby improving the robustness of vision models across various tasks and settings.

Contribution

It presents a novel approach to create 'robust objects' by exploiting model sensitivities, enhancing performance and robustness in computer vision applications.

Findings

Effective in standard benchmarks

Improves robustness in simulated robotics

Validates in real-world experiments

Abstract

We study a class of realistic computer vision settings wherein one can influence the design of the objects being recognized. We develop a framework that leverages this capability to significantly improve vision models' performance and robustness. This framework exploits the sensitivity of modern machine learning algorithms to input perturbations in order to design "robust objects," i.e., objects that are explicitly optimized to be confidently detected or classified. We demonstrate the efficacy of the framework on a wide variety of vision-based tasks ranging from standard benchmarks, to (in-simulation) robotics, to real-world experiments. Our code can be found at https://git.io/unadversarial .