iCartoonFace Challenge: Cartoon Face Detection and Recognition Competition

This document outlines the iCartoonFace challenge, a competition focused on advancing cartoon face detection and recognition technologies. The challenge addresses the need for large-scale, high-quality datasets to improve AI models' ability to identify cartoon characters, which currently lack sufficient data and suffer from noise issues.

The competition includes two tasks: Task A (Detection) requires algorithms to generate bounding boxes for cartoon faces in images, evaluated using mAP metrics. Task B (Recognition) involves sorting gallery images by similarity to a probe image, with ranking precision as the evaluation criterion. Both tasks emphasize model efficiency, with strict size and latency constraints.

The provided datasets are extensive and rigorously annotated. The detection dataset contains 50,000 training images with 91,163 faces, while the recognition dataset includes 389,678 images of 5,013 cartoon characters. Both datasets undergo quality checks to maintain error rates below 5%. The challenges highlight the complexity of cartoon faces, which exhibit high similarity and variability due to expressions, occlusions, and angles.

Participants must submit detailed methodologies and code for top-ranked teams. The competition aims to enhance academic and industrial capabilities in face recognition and detection through innovative solutions.