Go to ICLR 2026 Conference homepageActivation Matters: Adaptive and Activated Negative Labels for OOD Detection with Vision-Language Models
Keywords: OOD detection, vision-language models, negative labels, label activation
TL;DR: OOD detection can be significantly enhanced via activated negative labels.
Abstract: Out-of-distribution (OOD) detection aims to identify samples that deviate from in-distribution (ID).
One popular pipeline addresses this by introducing negative labels distant from ID classes and detecting OOD based on their distance to these labels.
However, such labels may present poor activation on OOD samples, failing to capture the OOD characteristics.
To address this, we propose an \underline{A}daptive and \underline{A}ctivated \underline{Neg}ative labels guided approach (AANeg), which dynamically evaluates activation levels across the corpus dataset and selects words with high activation responses as negative labels. Specifically, AANeg identifies high-confidence test images online and accumulates their assignment probabilities over the corpus to construct a label activation metric.
Such a metric leverages historical test samples to adaptively align with the test distribution, enabling the selection of distribution-adaptive activated negative labels.
By further exploring the activation information within the current testing batch, we introduce a more fine-grained, batch-adaptive variant.
To fully utilize label activation knowledge, we propose an activation-aware score function that emphasizes negative labels with stronger activations, boosting performance and enhancing its robustness to the label number.
Our approach is zero-shot, training-free, test-efficient, highly scalable, and grounded in theoretical justification.
Notably, on the large-scale ImageNet benchmark, AANeg significantly reduces the FPR95 from 17.5\% to 9.8\%.
Codes will be released.
Primary Area: alignment, fairness, safety, privacy, and societal considerations
Submission Number: 15715
Loading