The GAN Model for Three-Dimensional Animation Graphic Design Under Convolutional Neural Network

To increase design efficiency and model recognition accuracy, this study uses the enhanced generative adversarial network (GAN) model for three-dimensional (3D) animation graphic design. It entails using point clouds to visualize 3D and taking color pictures of 3D animated scenes from various angles. A convolutional neural network (CNN), earth-mover's distance, and the least squares method are used to enhance the GAN model and produce high-quality point cloud outputs. The study evaluates the effects of several upgraded models in producing 3D animation scene images and examines how well the enhanced GAN performs in visual design for 3D animation scenes. The results show that the GAN model improved by the deep CNN and label smoothing (LS) processing improves the image quality of the 3D animation scene with the increase in the number of iterations. Additionally, the model improved with the least squares method has a more stable training process and lower loss values compared to the model improved by the earth-mover's distance.