AdaSCALE: Adaptive Scaling for OOD Detection

The ability of the deep learning model to recognize when a sample falls outside its learned distribution is critical for safe and reliable deployment. Recent state-of-the-art out-of-distribution (OOD) detection methods leverage activation shaping to improve the separation between in-distribution (ID) and OOD inputs. These approaches resort to sample-specific scaling but apply a static percentile threshold across all samples regardless of their nature, resulting in suboptimal ID-OOD separability. In this work, we propose \textbf{AdaSCALE}, an adaptive scaling procedure that dynamically adjusts the percentile threshold based on a sample's estimated OODness. This estimation leverages our key observation: OOD samples exhibit significantly more pronounced activation shifts at high-magnitude activations under minor perturbation compared to ID samples. AdaSCALE enables stronger scaling for likely ID samples and weaker scaling for likely OOD samples, yielding highly separable energy scores. Our approach achieves state-of-the-art OOD detection performance, outperforming the latest rival OptFS by $\textbf{14.94}$% in near-OOD and $\textbf{21.28}$% in far-OOD datasets in average FPR@95 metric on the ImageNet-1k benchmark across eight diverse architectures.

AI models trained to recognize images can be confidently wrong when shown something unfamiliar. A medical model trained on common diseases, for instance, should flag a rare condition instead of misdiagnosing it. Detecting unfamiliar inputs reliably remains an open problem. Recent methods amplify the model's internal signals to help tell familiar inputs apart from unfamiliar ones. But, they use the same fixed rule for every input which limits how well the two can be separated. We propose AdaSCALE, which adjusts the amplification per input based on how unfamiliar it looks. The estimate comes from a simple observation: slightly nudging the pixels of an unfamiliar input shifts the model's strongest internal signals much more than it does for a familiar one.