Comparison of Classification Neural Network for Shape Feature in AI-driven Metaverse Application
Qimiao Zeng, Zhehao Zhuang · 2023
Image classification technology is critical to the success of Metaverse applications. The key challenges in big data image classification is finding an efficient and automated method to classify images. Currently, the most popular approach is to use neural network to train an image classifier. Although various neural network models are available for image classification, there is a lack of model selection methods for specific scenarios in AI-driven Metaverse application. To address this issue, we introduced three popular neural networks in the industry and applies them to problem of shape classification in the interdisciplinary field of ceramic culture digitalization technology. Then, we use a two-parameter optimization training scheme with multiple batches to obtain multiple comparable results. Finally, we determine the optimal neural network selection scheme for ceramic type classification by combining the accuracy, number of parameters, reasoning time and other factors with the requirements of Metaverse scenarios.