期刊
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021)
卷 -, 期 -, 页码 9264-9272出版社
IEEE
DOI: 10.1109/ICCV48922.2021.00915
关键词
-
资金
- European Institute of Innovation & Technology (EIT)
- H2020 EU project SPRING - Socially Pertinent Robots in Gerontological Healthcare
This paper introduces a novel approach for Novel Class Discovery (NCD) by using a Unified Objective function (UNO) to enhance synergy between supervised and unsupervised learning. By utilizing a multi-view self-labeling strategy to generate pseudo-labels, it achieves a single classification objective operating on both known and unknown classes.
In this paper, we study the problem of Novel Class Discovery (NCD). NCD aims at inferring novel object categories in an unlabeled set by leveraging from prior knowledge of a labeled set containing different, but related classes. Existing approaches tackle this problem by considering multiple objective functions, usually involving specialized loss terms for the labeled and the unlabeled samples respectively, and often requiring auxiliary regularization terms. In this paper we depart from this traditional scheme and introduce a UNified Objective function (UNO) for discovering novel classes, with the explicit purpose of favoring synergy between supervised and unsupervised learning. Using a multi-view self-labeling strategy, we generate pseudo-labels that can be treated homogeneously with ground truth labels. This leads to a single classification objective operating on both known and unknown classes. Despite its simplicity, UNO outperforms the state of the art by a significant margin on several benchmarks (approximate to+10% on CIFAR-100 and +8% on ImageNet). The project page is available at : https://ncd-uno.github.io.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据