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

2026-01-18

AI Summary

Deep neural networks often produce overconfident predictions on out-of-distribution (OOD) samples during inference. To address this issue, this work proposes DCAC, a training-free, test-time calibration module that maintains a separate cache for each known class to dynamically collect high-entropy samples and performs two-stage lightweight calibration by fusing visual features with prediction probabilities. DCAC introduces, for the first time, a class-aware caching mechanism that effectively captures intra-class visual similarities between OOD samples and known categories. The method is compatible with both unimodal and vision-language models. Experimental results demonstrate that DCAC significantly enhances the performance of various OOD detection approaches; for instance, when combined with ASH-S on ImageNet, it reduces the FPR95 metric by 6.55%.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Large Vision Models
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

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.

Problem

• out-of-distribution detection
• overconfident predictions
• deep neural networks
• OOD samples

Innovation

• Dynamic Class-Aware Cache
• Out-of-Distribution Detection
• Test-Time Calibration
• Training-Free
• Vision-Language Models