arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2404.03248cs.CV
收藏

学习用于分布外检测的可迁移负向提示

Learning Transferable Negative Prompts for Out-of-Distribution Detection

Beihang University(北京航空航天大学);Jiangxi Research Institute(江西研究院);Singapore Management University(新加坡管理大学)

Tianqi Li, Guansong Pang, Xiao Bai, Wenjun Miao, Jin Zheng

AI总结:

本文提出NegPrompt,仅用分布内数据学习可迁移到新类别的负向提示以界定ID/OOD边界,并在多个ImageNet基准上取得闭集与开放词汇场景下的SOTA OOD检测性能。

摘要:

现有的提示学习方法在分布外(OOD)检测方面已展现出一定能力,但在其训练过程中目标数据集缺少OOD图像,这会导致OOD图像与分布内(ID)类别之间出现错配,从而产生较高的假阳性率。为解决这一问题,我们提出了一种名为“NegPrompt”的新型OOD检测方法,用于学习一组负向提示,每个负向提示都表示给定类别标签的一种负面内涵,以刻画ID图像与OOD图像之间的边界。该方法仅使用ID数据学习此类负向提示,不依赖任何外部离群数据。此外,当前方法假设能够获得所有ID类别的样本,这使其在开放词汇学习场景中失效,因为推理阶段可能包含训练期间不存在的新ID类别。相比之下,我们学习到的负向提示可迁移到新的类别标签。在多个ImageNet基准上的实验表明,NegPrompt优于基于提示学习的最先进OOD检测方法,并在闭集和开放词汇分类场景的困难OOD检测中保持稳定领先。代码可在https://github.com/mala-lab/negprompt获取。

英文摘要:

Existing prompt learning methods have shown certain capabilities in Out-of-Distribution (OOD) detection, but the lack of OOD images in the target dataset in their training can lead to mismatches between OOD images and In-Distribution (ID) categories, resulting in a high false positive rate. To address this issue, we introduce a novel OOD detection method, named 'NegPrompt', to learn a set of negative prompts, each representing a negative connotation of a given class label, for delineating the boundaries between ID and OOD images. It learns such negative prompts with ID data only, without any reliance on external outlier data. Further, current methods assume the availability of samples of all ID classes, rendering them ineffective in open-vocabulary learning scenarios where the inference stage can contain novel ID classes not present during training. In contrast, our learned negative prompts are transferable to novel class labels. Experiments on various ImageNet benchmarks show that NegPrompt surpasses state-of-the-art prompt-learning-based OOD detection methods and maintains a consistent lead in hard OOD detection in closed- and open-vocabulary classification scenarios. Code is available at https://github.com/mala-lab/negprompt.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑