ArmorNet: Animated Cartoon Pornography Detection Using Transformer Network

Abstract

The management of inappropriate video content has become a critical issue in today's society, amplified by the internet's extensive accessibility. The control of sensitive material, such as explicit content and adult scenes, holds significant importance. Although advancements in machine learning and neural networks have enabled the creation of models to filter explicit content in movies, these methods might prove inadequate for efficiently filtering cartoon movies for children. This inadequacy stems from the differing criteria required for filtering content meant for children versus adults, with distinct parameters for age groups. To address this disparity, ArmorNet, a deep neural network model, has been proposed and compared against ResNet in a study that evaluated classification accuracy on Danbooru2020, a recognized dataset of animated cartoon images. This comparison highlights ArmorNet's superior performance and efficiency, despite the substantial computational resources needed for training. Notably, ArmorNet's lightweight design prioritizes classification accuracy, rendering it suitable for low-resource devices like mobile phones and gaming consoles.