Rethinking Out-of-Distribution Detection From a Human-Centric Perspective
This work addresses reliability and safety issues in deep neural networks for real-world applications, but it is incremental as it refines evaluation rather than introducing a new detection method.
The paper tackles the problem of out-of-distribution (OOD) detection by highlighting a discrepancy in conventional evaluation methods, which ignore risks from input-space shifts and misclassified images, and proposes a human-centric perspective to align detection with human expectations. It finds that simple baseline methods achieve comparable or better performance than recent approaches on 45 classifiers and 8 test datasets, suggesting overestimation of past developments.
Out-Of-Distribution (OOD) detection has received broad attention over the years, aiming to ensure the reliability and safety of deep neural networks (DNNs) in real-world scenarios by rejecting incorrect predictions. However, we notice a discrepancy between the conventional evaluation vs. the essential purpose of OOD detection. On the one hand, the conventional evaluation exclusively considers risks caused by label-space distribution shifts while ignoring the risks from input-space distribution shifts. On the other hand, the conventional evaluation reward detection methods for not rejecting the misclassified image in the validation dataset. However, the misclassified image can also cause risks and should be rejected. We appeal to rethink OOD detection from a human-centric perspective, that a proper detection method should reject the case that the deep model's prediction mismatches the human expectations and adopt the case that the deep model's prediction meets the human expectations. We propose a human-centric evaluation and conduct extensive experiments on 45 classifiers and 8 test datasets. We find that the simple baseline OOD detection method can achieve comparable and even better performance than the recently proposed methods, which means that the development in OOD detection in the past years may be overestimated. Additionally, our experiments demonstrate that model selection is non-trivial for OOD detection and should be considered as an integral of the proposed method, which differs from the claim in existing works that proposed methods are universal across different models.