Abstract
Cartoons receive increasingly attention and have a huge global market.Cartoon person recognition has a wealth of application scenarios. However,there is no large and high quality dataset for cartoon person recognition. Itlimit the development of recognition algorithms. In this paper, we propose thefirst large unconstrained cartoon database called iCartoonFace. We havereleased the dataset publicly available to promote cartoon person recognitionresearch\footnote{The dataset can be applied by sending email [email protected]}. The dataset contains 68,312 images of 2,639 identities.The dataset consists of persons which come from cartoon videos. The samples areextracted from public available images on website and online videos on iQiYicompany. All images pass through a careful manual annotation process. Weevaluated the state-of-the-art image classification and face recognitionalgorithms on the iCartoonFace dataset as a baseline. A dataset fusion methodwhich utilize face feature to improve the performance of cartoon recognitiontask is proposed. Experimental performance show that the performance ofbaseline models much worse than human performance. The proposed dataset fusionmethod achieves a 2\% improvement over the baseline model. In a word,state-of-the-art algorithms for classification and recognition are far frombeing perfect for unconstrained cartoon person recognition.