Cartoon Face Recognition: Introducing the iCartoonFace Benchmark Dataset

iQIYI's recent paper "Cartoon Face Recognition: A Benchmark Dataset" has been accepted by the ACM Multimedia conference, the premier international venue for multimedia research.

The paper introduces iCartoonFace, the largest manually annotated cartoon face dataset worldwide, containing over 5,000 distinct cartoon characters and more than 400,000 high‑quality real‑scene images. The dataset aims to accelerate research in cartoon character recognition, a field that lags behind real‑face recognition despite growing demand from animation, video editing, image search, and advertising.

iCartoonFace addresses the challenges of cartoon data, such as high intra‑class variance and low inter‑class differences, and provides a semi‑automatic data‑collection pipeline and a multi‑person training framework that combine cartoon and real‑person images.

The dataset is already being applied in iQIYI's AI services, including automatic cartoon material collection, theme‑based video mash‑ups, and the "奇观" feature that lets users identify cartoon characters in anime works with a single click.

iQIYI has also organized several competitions around the dataset, collaborating with IJCAI PRICAI 2020 to launch the first large‑scale cartoon face recognition contest in China, attracting nearly 500 teams from top universities and leading tech companies.

For the full paper, see https://dl.acm.org/doi/abs/10.1145/3394171.3413726 .