DCAC: Dynamic Class-Aware Cache Creates Stronger Out-of-Distribution Detectors

TL;DR

DCAC introduces a class-aware cache mechanism that improves out-of-distribution detection by calibrating predictions at test time, significantly reducing false positives across various models and benchmarks.

Contribution

The paper proposes DCAC, a training-free, test-time calibration method that maintains class-specific caches to enhance OOD detection performance.

Findings

DCAC reduces FPR95 by 6.55% with ASH-S on ImageNet.

It improves OOD detection across multiple benchmarks.

DCAC is lightweight and compatible with various models.

Abstract

Out-of-distribution (OOD) detection remains a fundamental challenge for deep neural networks, particularly due to overconfident predictions on unseen OOD samples during testing. We reveal a key insight: OOD samples predicted as the same class, or given high probabilities for it, are visually more similar to each other than to the true in-distribution (ID) samples. Motivated by this class-specific observation, we propose DCAC (Dynamic Class-Aware Cache), a training-free, test-time calibration module that maintains separate caches for each ID class to collect high-entropy samples and calibrate the raw predictions of input samples. DCAC leverages cached visual features and predicted probabilities through a lightweight two-layer module to mitigate overconfident predictions on OOD samples. This module can be seamlessly integrated with various existing OOD detection methods across both unimodal and vision-language models while introducing minimal computational overhead. Extensive experiments on multiple OOD benchmarks demonstrate that DCAC significantly enhances existing methods, achieving substantial improvements, i.e., reducing FPR95 by 6.55% when integrated with ASH-S on ImageNet OOD benchmark.