InspectorNet: Transformer network for violence detection in animated cartoon | ||||
Engineering Research Journal (Shoubra) | ||||
Volume 52, Issue 2, April 2023, Page 114-119 PDF (799.46 K) | ||||
Document Type: Research articles | ||||
DOI: 10.21608/erjsh.2023.181713.1119 | ||||
View on SCiNiTO | ||||
Authors | ||||
Mahmoud Mohammed Taha 1; Abdelwahab Kamel Alsammak 1; Ahmed Bayiomy Zaky 1, 2 | ||||
1Electrical Engineering Department, Faculty of Engineering at Shoubra, Benha University. | ||||
2Egypt japan university of science and technology computer science and information technology programs (CSIT) | ||||
Abstract | ||||
InspectorNet is a convolutional neural network based on transformer deep learning techniques, which is designed to address some of the limitations of current state-of-the-art artificial neural networks (ANN) models. The paper compares the performance of InspectorNet against a commonly used neural network in image classification, ResNet [1], on the Danbooru2020 dataset of animated cartoon images with a variable number of classes. The comparison shows that while both networks require significant computing resources for training, InspectorNet demonstrates better classification performance in certain test situations. The paper also highlights that with the increasing access to the internet, it is important to control the dissemination of sensitive content such as violence, but current neural networks may not be as effective in filtering cartoon movies aimed at children as the filters for these movies are different from those for adult movies. InspectorNet also has a compact architecture than many modern networks, such as ResNet, which results in better performance on low-resource devices | ||||
Keywords | ||||
violence filtering; transformer; image classification | ||||
Statistics Article View: 278 PDF Download: 191 |
||||