IoUCert: Robustness Verification for Anchor-based Object Detectors

TL;DR

IoUCert is a new formal verification framework that enables robustness guarantees for anchor-based object detectors by directly bounding IoU metrics, overcoming previous limitations due to complex transformations.

Contribution

We introduce IoUCert, a novel verification method that directly bounds IoU in anchor-based detectors, enabling robustness guarantees for models like SSD and YOLO variants.

Findings

First to verify robustness of realistic anchor-based detectors

Enables bounds on IoU for models like SSD, YOLOv2, YOLOv3

Demonstrates robustness against input perturbations

Abstract

While formal robustness verification has seen significant success in image classification, scaling these guarantees to object detection remains notoriously difficult due to complex non-linear coordinate transformations and Intersection-over-Union (IoU) metrics. We introduce IoUCert, a novel formal verification framework designed specifically to overcome these bottlenecks in foundational anchor-based object detection architectures. Focusing on the object localisation component in single-object settings, we propose a coordinate transformation that enables our algorithm to circumvent precision-degrading relaxations of non-linear box prediction functions. This allows us to optimise bounds directly with respect to the anchor box offsets which enables a novel Interval Bound Propagation method that derives optimal IoU bounds. We demonstrate that our method enables, for the first time, the robustness verification of realistic, anchor-based models including SSD, YOLOv2, and YOLOv3 variants against various input perturbations.