MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition
In this paper, we design a benchmark task and provide the associated datasets for recognizing face images and link them to corresponding entity keys in a knowledge base. More specifically, we propose a benchmark task to recognize one million celebrities f
- PDF / 2,701,714 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 94 Downloads / 197 Views
Abstract. In this paper, we design a benchmark task and provide the associated datasets for recognizing face images and link them to corresponding entity keys in a knowledge base. More specifically, we propose a benchmark task to recognize one million celebrities from their face images, by using all the possibly collected face images of this individual on the web as training data. The rich information provided by the knowledge base helps to conduct disambiguation and improve the recognition accuracy, and contributes to various real-world applications, such as image captioning and news video analysis. Associated with this task, we design and provide concrete measurement set, evaluation protocol, as well as training data. We also present in details our experiment setup and report promising baseline results. Our benchmark task could lead to one of the largest classification problems in computer vision. To the best of our knowledge, our training dataset, which contains 10M images in version 1, is the largest publicly available one in the world. Keywords: Face recognition · Large scale · Benchmark · Training data · Celebrity recognition · Knowledge base
1
Introduction
In this paper, we design a benchmark task as to recognize one million celebrities from their face images and identify them by linking to the unique entity keys in a knowledge base. We also construct associated datasets to train and test for this benchmark task. Our paper is mainly to close the following two gaps in current face recognition, as reported in [1]. First, there has not been enough effort in determining the identity of a person from a face image with disambiguation, especially at the web scale. The current face identification task mainly focuses on finding similar images (in terms of certain types of distance metric) for the input image, rather than answering questions such as “who is in the image?” and “if it is Anne in the image, which Anne?”. This lacks an important step of “recognizing”. The second gap is about the scale. The publicly available datasets are much smaller than that being used privately in industry, such as Facebook [2,3] and Google [4], as summarized in Table 1. Though the research in face recognition highly desires large datasets consisting of many distinct people, such c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part III, LNCS 9907, pp. 87–102, 2016. DOI: 10.1007/978-3-319-46487-9 6
88
Y. Guo et al.
large dataset is not easily or publicly accessible to most researchers. This greatly limits the contributions from research groups, especially in academia. Our benchmark task has the following properties. First, we define our face recognition as to determine the identity of a person from his/her face images. More specifically, we introduce a knowledge base into face recognition, since the recent advance in knowledge bases has demonstrated incredible capability of providing accurate identifiers and rich properties for celebrities. Examples include Satori knowledge graph in Microsoft and “freebas
Data Loading...
