Semantic Segmentation of Balinese Dance Agem With VGG19-UNET Method

This research examines the application of the Semantic Segmentation method using VGG19-UNET architecture to map agem movements in Balinese dance. U-NET was chosen for its ability to improve the accuracy and efficiency of semantic segmentation, which aims to break down agem movements into smaller parts that are easier to learn. This study compares the performance of Adam and Nadam optimization in training the VGG19-UNET model. The best results show an accuracy of $\mathbf{9 6 . 9 9 \%}$ for Adam optimization and $\mathbf{9 6 . 9 9 \%}$ for Nadam optimization, with losses of $0.89 \%$ and $0.93 \%$, respectively, and validation losses of $1.33 \%$ and $1.35 \%$. This research successfully proves that the developed model can perform semantic segmentation of agem movements in Balinese dance with a very good level of accuracy and prediction. This result is expected to contribute as a learning tool for Balinese dance and encourage the younger generation of Bali to preserve their cultural heritage by utilizing modern technology.