Map scene generation framework based on language models

The design of map scenes is significant in the virtual reality and various other domains, gradually emerging as a key area of research. However, conventional coding-dependent approaches still demand high levels of expertise from designers and are associated with substantial design costs. Thus, in this paper, we introduce a map scene design framework by leveraging language models to streamline the process for users engaged in the map scene design. Specifically, our approach relies solely on natural language text to manipulate the map engine and fine-tune smaller models within the mapping domain, thereby eliminating the need for users to undertake secondary development. We illustrate our framework has been tested in 3D map scene design, with results indicating strong accuracy and efficient drawing capabilities.