SimLabel: Consistency-Guided OOD Detection with Pretrained Vision-Language Models
Keywords: Out-of-distribution detection, Vision-Language Models
Abstract: Detecting out-of-distribution (OOD) data is crucial in real-world machine learning applications to prevent severe errors, particularly in safety-critical domains. Existing methods often leverage language information from vision-language models (VLMs) to enhance OOD detection by improving confidence estimation through rich class-wise text information. However, those methods primarily focus on obtaining OOD scores based on the similarity of the new sample to each in-distribution (ID) class, overlooking the OOD scores to a group of similar classes. We assume that an ID sample should consistently receive high similarity score across similar ID classes. This paper investigates the ability of image-text comprehension among different semantic-related ID labels in VLMs and proposes a novel post-hoc strategy called SimLabel. SimLabel enhances the separability between ID and OOD samples by establishing a more robust image-class similarity metric that considers consistency over a set of similar class labels. Extensive experiments demonstrate the superior performance of SimLabel across various zero-shot OOD detection benchmarks, underscoring its efficacy in achieving robust OOD detection.
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: 6402
Loading