Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection
This work addresses the problem of safely deploying models in open-world environments for computer vision practitioners, but it is incremental as it builds on existing post-hoc detection methods.
The paper tackles out-of-distribution detection in deep computer vision by proposing a post-hoc method that leverages perturbation robustness, resulting in a more than 10% reduction in FPR@95 compared to state-of-the-art methods on a CIFAR-10 model.
Out-of-distribution (OOD) detection is the task of identifying inputs that deviate from the training data distribution. This capability is essential for safely deploying deep computer vision models in open-world environments. In this work, we propose a post-hoc method, Perturbation-Rectified OOD detection (PRO), based on the insight that prediction confidence for OOD inputs is more susceptible to reduction under perturbation than in-distribution (IND) inputs. Based on the observation, we propose an adversarial score function that searches for the local minimum scores near the original inputs by applying gradient descent. This procedure enhances the separability between IND and OOD samples. Importantly, the approach improves OOD detection performance without complex modifications to the underlying model architectures. We conduct extensive experiments using the OpenOOD benchmark~\cite{yang2022openood}. Our approach further pushes the limit of softmax-based OOD detection and is the leading post-hoc method for small-scale models. On a CIFAR-10 model with adversarial training, PRO effectively detects near-OOD inputs, achieving a reduction of more than 10\% on FPR@95 compared to state-of-the-art methods.