arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2004.10862cs.CV
收藏

对象学习的持续学习

Continual Learning of Object Instances

University of Amsterdam(阿姆斯特丹大学);Prosus

Kishan Parshotam, Mert Kilickaya

摘要:

我们提出了持续的实例学习-一种将持续学习的概念应用于区分相同对象类别的实例的任务的方法。我们特别关注汽车对象,并通过度量学习逐步学习区分汽车实例。我们开始我们的论文通过评估当前的技术。在确定现有方法中灾难性遗忘是明显的之后,我们提出了两种补救措施。首先,我们通过归一化交叉熵来规范度量学习。其次,我们用合成数据传输来增强现有模型。我们在三个大规模数据集上进行了广泛的实验,使用两种不同的架构用于五种不同的持续学习方法,结果表明,归一化交叉熵和合成转移导致现有技术中的遗忘较少。

英文摘要:

We propose continual instance learning - a method that applies the concept of continual learning to the task of distinguishing instances of the same object category. We specifically focus on the car object, and incrementally learn to distinguish car instances from each other with metric learning. We begin our paper by evaluating current techniques. Establishing that catastrophic forgetting is evident in existing methods, we then propose two remedies. Firstly, we regularise metric learning via Normalised Cross-Entropy. Secondly, we augment existing models with synthetic data transfer. Our extensive experiments on three large-scale datasets, using two different architectures for five different continual learning methods, reveal that Normalised cross-entropy and synthetic transfer leads to less forgetting in existing techniques.

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。

↑