arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.02748cs.CVcs.AIcs.LGeess.IV
收藏

CLAD:一种基于对比学习的背景去偏方法

CLAD: A Contrastive Learning based Approach for Background Debiasing

Ke Wang, Harshitha Machiraju, Oh-Hyeon Choung, Michael Herzog, Pascal Frossard

摘要:

卷积神经网络(CNN)在多个视觉任务中取得了超人的性能,特别是在图像分类方面。然而,与人类不同的是,CNN利用虚假特征(如背景信息)来做出决策。这种趋势在鲁棒性或弱泛化性能方面产生了不同的问题。通过我们的工作,我们引入了一种基于对比学习的方法(CLAD)来减轻CNN中的背景偏差。CLAD鼓励语义关注对象前景,并惩罚从无关背景中学习特征。我们的方法还介绍了一种有效的方法来采样负样本。我们在背景挑战数据集上获得了最先进的结果,以4.1%的利润率超过了之前的基准。我们的论文展示了CLAD如何作为一个概念证明,用于消除虚假特征,如背景和纹理(在补充材料中)。

英文摘要:

Convolutional neural networks (CNNs) have achieved superhuman performance in multiple vision tasks, especially image classification. However, unlike humans, CNNs leverage spurious features, such as background information to make decisions. This tendency creates different problems in terms of robustness or weak generalization performance. Through our work, we introduce a contrastive learning-based approach (CLAD) to mitigate the background bias in CNNs. CLAD encourages semantic focus on object foregrounds and penalizes learning features from irrelavant backgrounds. Our method also introduces an efficient way of sampling negative samples. We achieve state-of-the-art results on the Background Challenge dataset, outperforming the previous benchmark with a margin of 4.1\%. Our paper shows how CLAD serves as a proof of concept for debiasing of spurious features, such as background and texture (in supplementary material).

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。

↑