Abstract:In the dynamic realms of machine learning and deep learning, the robustness and reliability of models are paramount, especially in critical real-world applications. A fundamental challenge in this sphere is managing Out-of-Distribution (OOD) samples, significantly increasing the risks of model misclassification and uncertainty. Our work addresses this challenge by enhancing the detection and management of OOD samples in neural networks. We introduce OOD-R (Out-of-Distribution-Rectified), a meticulously curated collection of open-source datasets with enhanced noise reduction properties. In-Distribution (ID) noise in existing OOD datasets can lead to inaccurate evaluation of detection algorithms. Recognizing this, OOD-R incorporates noise filtering technologies to refine the datasets, ensuring a more accurate and reliable evaluation of OOD detection algorithms. This approach not only improves the overall quality of data but also aids in better distinguishing between OOD and ID samples, resulting in up to a 2.5\% improvement in model accuracy and a minimum 3.2\% reduction in false positives. Furthermore, we present ActFun, an innovative method that fine-tunes the model's response to diverse inputs, thereby improving the stability of feature extraction and minimizing specificity issues. ActFun addresses the common problem of model overconfidence in OOD detection by strategically reducing the influence of hidden units, which enhances the model's capability to estimate OOD uncertainty more accurately. Implementing ActFun in the OOD-R dataset has led to significant performance enhancements, including an 18.42\% increase in AUROC of the GradNorm method and a 16.93\% decrease in FPR95 of the Energy method. Overall, our research not only advances the methodologies in OOD detection but also emphasizes the importance of dataset integrity for accurate algorithm evaluation.

FodFoM: Fake Outlier Data by Foundation Models Creates Stronger Visual Out-of-Distribution Detector

TagFog: Textual Anchor Guidance and Fake Outlier Generation for Visual Out-of-Distribution Detection

Can OOD Object Detectors Learn from Foundation Models?

OAL: Enhancing OOD Detection Using Latent Diffusion

Rethinking Out-of-distribution (OOD) Detection: Masked Image Modeling is All You Need

Detecting Out-of-Distribution Examples Via Class-Conditional Impressions Reappearing

Advancing Out-of-Distribution Detection through Data Purification and Dynamic Activation Function Design

Out-of-Distribution Detection with Prototypical Outlier Proxy

Out-of-Distribution Detection with Virtual Outlier Smoothing

Going Beyond Conventional OOD Detection

Enhancing Out-of-Distribution Detection with Multitesting-based Layer-wise Feature Fusion

Dream the Impossible: Outlier Imagination with Diffusion Models

Zero-Shot Out-of-Distribution Detection with Outlier Label Exposure

Out-of-Distribution Identification: Let Detector Tell Which I Am Not Sure.

The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection

TagOOD: A Novel Approach to Out-of-Distribution Detection via Vision-Language Representations and Class Center Learning

Image Background Serves as Good Proxy for Out-of-distribution Data

RONF: Reliable Outlier Synthesis under Noisy Feature Space for Out-of-Distribution Detection

When an extra rejection class meets out-of-distribution detection in long-tailed image classification

Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection

Out-of-distribution Detection with Diffusion-based Neighborhood