A dance movement quality evaluation model using transformer encoder and convolutional neural network

The purpose of this study is to put forward a new evaluation model of dance movement quality to deal with the subjectivity and inconsistency in traditional evaluation methods. In view of the complexity and diversity of dance art and the widespread popularity of dance videos on social media, it is particularly urgent to develop an automatic and efficient tool for evaluating the quality of dance movements. Therefore, this study puts forward the Transformer Convolutional Neural Network with Dynamic and Static Streams (TransCNN-DSSS) model, which combines the analysis of dynamic flow and static flow, and makes use of the advantages of Transformer and Convolutional Neural Network (CNN) to deeply analyze and evaluate the dance movements. The core of the model is Quality Score Decoupling (QSD), which decouples and weights different quality dimensions of dance movements through attention mechanism, such as accuracy, fluency and expressiveness. Score Prediction module (SPM) uses Transformer network to further process the fused features, and outputs the final evaluation score through the full connection layer. In the experimental part, the TransCNN-DSSS model is trained and tested on the marked dance movement dataset. The performance of the model is evaluated by accuracy, recall and F1 score. The results show that the model has achieved 90% accuracy, 89% recall and F1 score of 0.90 in the task of evaluating the quality of dance movements. These results prove the effectiveness and reliability of the model. In addition, the adaptability test of the model in different dance styles also shows good generalization ability. The research contribution of this study is to put forward a new evaluation model of dance movement quality, which provides an objective and automatic evaluation tool for dance teaching, competition scoring and fans.