The issues that are encountered in intelligent layout generation of animation scenes include lack of semantic knowledge and unreasonable layout of the scene components leading to lack of logical consistency and visual aesthetics in the generated scenes. In the current paper, an intelligent layout generation technique is suggested, consisting of semantic understanding and Generative Adversarial Network (GAN), allowing automatic semantic-based generation of animation scenes. The approach consists of: (1) building a semantic parsing module to infer the core objects and their spatial interactions in the scene description based on a natural language processing model; (2) building a semantically constrained layout generator to predict the relative position and proportion of objects by a position mapping network; (3) employing a multi-scale loss function to optimize the generated outputs; (4) a dual discriminator GAN architecture, one to judge the consistency of the entire scene, and the other to judge the rationality of the layout. The experiment was carried out using a self-constructed animation scene corpus and the test set comprised of 1200 sets of semantic description and layout comparison data. The results show that the method presented in this paper achieves an average of 92.6 % in terms of layout rationality index, with a generation time of only 0.77–0.86 seconds. In 8 out of 10 sets of data, the matching degree exceeds 93%. This method provides an effective path for intelligent animation production.
Liang et al. (Thu,) studied this question.