arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1707.01700cs.CVcs.AIcs.LG
收藏

CNN特性在无监督分类方面也很出色

CNN features are also great at unsupervised classification

Laboratoire des Sciences de l’Information et des Systèmes (CNRS UMR 7296)(信息与系统科学实验室(法国国家科学研究中心 UMR 7296));Arts et Métiers ParisTech(巴黎高等工艺学院)

Joris Guérin, Olivier Gibaru, Stéphane Thiery, Eric Nyiri

摘要:

本文旨在深入了解深度CNN特征对无监督问题的可转移性。我们研究了不同预训练的CNN特征提取器对图像集聚类问题的影响,以进行对象分类和细粒度分类。我们提出了一个相当简单的管道,使用在ImageNet上预训练的CNN和经典的聚类算法对图像集进行分类,结合深度特征提取。这种方法相比,国家的最先进的图像聚类算法,并提供更好的结果。这些结果加强了这样一种信念,即在大型数据集上对深度CNN进行监督训练,具有很大的类别变化性,即使对于无监督任务,也比大多数精心设计的工程方法提取更好的特征。我们还验证了我们的机器人应用程序的方法,包括在分类和存储对象智能的基础上聚类。

英文摘要:

This paper aims at providing insight on the transferability of deep CNN features to unsupervised problems. We study the impact of different pretrained CNN feature extractors on the problem of image set clustering for object classification as well as fine-grained classification. We propose a rather straightforward pipeline combining deep-feature extraction using a CNN pretrained on ImageNet and a classic clustering algorithm to classify sets of images. This approach is compared to state-of-the-art algorithms in image-clustering and provides better results. These results strengthen the belief that supervised training of deep CNN on large datasets, with a large variability of classes, extracts better features than most carefully designed engineering approaches, even for unsupervised tasks. We also validate our approach on a robotic application, consisting in sorting and storing objects smartly based on clustering.

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。

↑