GRIC: General Representation and Informative Content for Enhanced Out-of-Distribution Detection
Keywords: Out-of-Distribution Detection
Abstract: Out-of-distribution (OOD) detection is crucial for ensuring the robustness of machine learning models in open-world scenarios by identifying inputs from unknown classes. Vision-language models like CLIP have enabled zero-shot OOD detection without requiring labels or training on in-distribution (ID) data. However, current approaches are limited by their dependence on \textit{closed-set text-based labels} and \textit{full image feature representations}, constraining CLIP’s capacity to generalize across diverse labels. In this work, we propose GRIC, a novel method that improves zero-shot multi-modal OOD detection by leveraging two key insights: (1) OOD detection is driven by general ID representations rather than class-specific features, and (2) large language models (LLMs) can enrich the model’s understanding of ID data and simulate potential OOD scenarios without actual OOD samples. GRIC is simple yet highly effective, reducing the false positive rate at $95\%$ recall (FPR95) by up to $19\%$, significantly surpassing state-of-the-art methods.
Supplementary Material: pdf
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1665
Loading