DocFigure: A Dataset for Scientific Document Figure Classification
K. V. Jobin, Ajoy Mondal, C. V. Jawahar · 2019
Document figure classification (DFC) is an important stage of a document figure understanding system. The design of a DFC system required a well defined figure categories and dataset. To the best of the author's knowledge, the existing datasets related to classification of figures in the document images are limited with respect to their size and categories [1]-[3]. In this paper, we introduce a scientific figure classification dataset, named as DocFigure. The dataset consists of 33K annotated figures of 28 different categories present in the document images which correspond to scientific articles published in the CVPR, ECCV, ICCV, etc. conferences in last several years. Manual annotation of such a large number (33K) of figures is time consuming and cost ineffective. In this article, we design a web based annotation tool which can efficiently assign category labels to large number of figures with the minimum efforts of human annotators. To benchmark our generated dataset on classification task, we propose three baseline classification techniques using deep feature, deep texture feature and combination of both. In our analysis, we found that the combination of both deep feature and deep texture feature is more effective for document figure classification task than the individual features. The dataset and the code are publicly available at https://researchweb.iiit.ac.in/~jobin.kv/projects/.