Towards lifelong object recognition: A dataset and benchmark. (October 2022)

• Record Type: Journal Article
• Authors: Lan, Chuanlin; Feng, Fan; Liu, Qi; She, Qi; Yang, Qihan; Hao, Xinyue; Mashkin, Ivan; Kei, Ka Shun; Qiang, Dong; Lomonaco, Vincenzo; Shi, Xuesong; Wang, Zhengwei; Guo, Yao; Zhang, Yimin; Qiao, Fei; Chan, Rosa H.M.
• Is Part Of: Pattern recognition. Volume 130(2022)
• Publication Date: 2022-10
• Subjects: Robotic vision -- Continual learning -- Lifelong learning -- Object recognition; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108819
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We release OpenLORIS-Object, which is the first real-world lifelong learning dataset for robotic vision with quantifiable environmental factors. We provide a general benchmark for evaluating lifelong learning capabilities on three continual learning scenarios. We analyze 9 SOTA algorithms tested on our dataset, finding that existing algorithms are insufficient to solve the task in a reallifeenvironment. Abstract: Lifelong learning algorithms aim to enable robots to handle open-set and detrimental conditions, and yet there is a lack of adequate datasets with diverse factors for benchmarking. In this work, we constructed and released a lifelong learning robotic vision dataset, OpenLORIS-Object. This dataset was collected by RGB-D camera capturing dynamic environment in daily life scenarios with diverse factors, including illumination, occlusion, object pixel size and clutter, of quantified difficulty levels. To the best of our knowledge, this is an unique real-world dataset for robotic vision with independent and quantifiable environmental factors, which are currently unaccounted for in other lifelong learning datasets such as CORe50 and NICO. We tested 9 state-of-the-art algorithms with 4 evaluation metrics over the dataset in Domain Incremental Learning, Task Incremental Learning, and Class Incremental Learning scenarios. The results demonstrate that these existing algorithms are insufficient to handle lifelong learning task in dynamic environments. Our dataset and benchmarks are now publicly available at this website . 2