Based on transfer learning and graph neural network for animated clothing element recognition

Animation costume element recognition has become a major computer vision research area given the fast growth of the animation sector. Conventions for costume recognition include issues including significant stylistic variations and little annotating data. This work suggests a framework for animation costume element detection based on transfer learning (TL) and graph neural network (GNN) in order to solve these difficulties AnimCloth-TL-GNN. First, the framework models the spatial relationship between costume elements by the TL module; then, the GNN module models the knowledge from the source domain to the target domain; finally, the fusion module combines the features of both to improve the recognition effect of the model. Strong advantages in the animated costume element recognition job are shown by the experimental findings of the AnimCloth-TL-GNN framework, thereby enhancing the accuracy and has great generalising capacity and robustness.